AI and the Net • Buyer’s GuideAI and the Net • Buyer’s Guide

Neural NetworksNeural Networks

18.318.3

Where Intelligent TechnologyWhere Intelligent TechnologyMeets the Real WorldMeets the Real Worldwww.pcai.comwww.pcai.com

Also:Also:Business Applications,Business Applications,

Data Mining,Data Mining,Data Modeling,Data Modeling,

Decision Support,Decision Support,Intelligent Search Tools,Intelligent Search Tools,

Intelligent Tools,Intelligent Tools,Robotics,Robotics,

GeneralGeneralAnnouncements,Announcements,AI Conferences, AI Conferences,

Training,Training,Fuzzy Logic,Fuzzy Logic,

Fuzzy SQL,Fuzzy SQL,Intelligent Process Control,Intelligent Process Control,

Neural Networks,Neural Networks,SpeechSpeech

Recognition,Recognition,Text Mining,Text Mining,

Web UtilitiesWeb Utilities

Neural NetworksNeural Networks

Indexed vs. UnindexedIndexed vs. UnindexedSearching: From SecuritySearching: From SecurityClassifications to ForensicsClassifications to Forensics

Neural Nets and ScientificNeural Nets and ScientificDiscovery: A Match Made in AIDiscovery: A Match Made in AIHeavenHeaven

The Visual Development ofThe Visual Development ofRule-Based SystemsRule-Based Systems

Protégé, Ontology andProtégé, Ontology andKnoweldge Acquisition:Knoweldge Acquisition:Knowledge Representation, Knowledge Representation, the Foundation of Intelligentthe Foundation of IntelligentSystemsSystems

AI and the Net • Buyer’s GuideAI and the Net • Buyer’s Guide

Where Intelligent Technology Meets the Real World

Theme: Neural Networks

Volume 18, Number 3

Table of Contents18 Indexed vs.Unindexed Searching: FromSecurity Classifications toForensicsElizabeth Thede discusses the dif-ferences between indexed andunindexed searching and when touse which technique.

24 Neural Nets andScientific Discovery: AMatch Made in AI HeavenIlana Marks presents some of theways that neural network technolo-gies are being used to expedite thescientific research process and pro-vide valuable insights that mightotherwise be overlooked.

29 The VisualDevelopment of Rule-BasedSystemsCharles Langley and CliveSpenser detail the creation of arule-based system using visual rep-resentations of rules rather than tra-ditional text-based representations.

37 Protégé, Ontologyand Knowledge Acquisition:Knowledge Representation,the Foundation ofIntelligent SystemsTerry Hengl discusses a knowl-edge-base application developmenttool called Protégé and how it usesontologies to define terms and con-cepts in useful ways.

58 Book Zone*Applying UML*Text Mining: Predictive Methods forAnalyzing Unstructured Information*Spoken Dialogue Technology:Towards the Conversational UserInterface*Fuzzy Control of Queuing Systems*Data Modeling Essentials, ThirdEdition*Requirements Engineering*Object-Oriented ConstructionHandbook: Developing Application-Oriented Software with the Tools &Materials Approach

Ilana Marks

62 Buyer’s GuideFuzzy Logic, Fuzzy SQL, IntelligentProcess Control, Neural Networks,Speech Recognition, Text Mining, WebUtilities

7 Editorial

8 Product Update BusinessApplications, Data Mining andModeling, Decision Support,Intelligent Search Tools, IntelligentTools, Modeling, Robotics,Announcements, AI Conferences,Training

49 AI-Q Crossword PuzzleNeural Networks, Fuzzy Logic, andSpeech Recognition

52 AI and the NetVirtual Museum Education Assistant,Call Center Computers, Nose Mouse,Network of Robotic Telescopes, andmore.

Ilana Marks

75 Classifieds and Recruiting

Page 24

Page 29

Columns

Regular Features

PC AI 5 18.3

Multiple PSM – Domain-IndependentPSM and automated form generationDefined three classes of ontology:• Domain, Method and Application

1986 1988 1990 1992 1994 1996 1998 2000 2002

Opal

Protégé-I

Protégé-II

Protégé/Win

Protégé-2000

Pre-Protégé

Single PSM - Episodic Skeletal -Plan Refinement (ESPR)PSM defines semanticsKnowledge bases are problem-specificBased on Xerox LISP Machine

Integrated Tool SetBased on MS Windows

Includes OKBC Knowledge ModelRuns under Java VM

Translated expert’s case entry into Oncocin’sinternal representation• Structural domain concepts,• oncology protocols• Case data exercise the expert system decision capability

Page 37

PC AI 6 18.3

PCPC AIAI Volume 17Volume 17now on CDnow on CD

6 Issues in both HMTL and PDFIssues 17.1 - 17.6 (2003-2004)

Additional updates and bonus video not inthe original issues.

All web issues reformatted with PC AINavigation Bar

Higher Quality PDF’s

Over 70 articles, columns and puzzles

Sample articles include:The Social Life of BooksThe DARPA Challenge

Searching Beyond the Tower of BabelAI and the Law

and much, much more.

Topics included:Data Mining, Robotics, Business Rules,

Genetic Algorithms, Prolog, LISP, BusinessForecasting, Case-Based Reasoning,

Knowledge Management, Agents, FuzzyLogic, Intelligent Tutoring, Expert Systems,

Logic Programming, Modeling andSimulation, Searching, Pattern Matching,

Natural Language Processing, etc.

Buy PC AI Volume 17 on CD

For Paid Subscribers - $14 each CD Non-Paid Subscribers - $32 each CD

US Postage - $3.00 for postage per CDForeign Postage - $5.00 per CD)

(Quantity Discounts are available)

Order online atwww.pcai.com/store

orContact PC AI directly

(602) 971-1869

Also Available: Volume 16 CDIssues 16.1 - 16.6 (2002-2003)PLUS a special bonus issue - 15.6 - AI to Combat Terrorism

Where Intelligent TechnologyMeets the Real World

PublisherTerry Hengl

Senior EditorDon Barker

WebmasterIlana Marks

Columnists/EditorsElisa HicksIlana Marks

Casper GoldbergPaul E. Grayson

ContributorsTerry Hengl

Charles LangleyIlana Marks

Clive SpenserElizabeth Thede

Layout and TypographyMichael Wiederhold

Graphic Design and IllustrationLaurens Watson

Vice President of MarketingRobin Okun

Editorial AssistantLauren Dana

Please direct all editorial andadvertising inquiries to:

PC AIPO Box 30130

Phoenix, AZ 85046-0130(602) 971-1869info@pcai.comwww.pcai.com

Subscription rates are $25 for oneyear (6 issues); $40 for two years

(12 issues)

Entire contents copyright ©2004 byKnowledge Technology, Inc., unless

otherwise noted.Authorization to photocopy itemspublished in PC AI Magazine for

internal or personal use, or the internalor personal use of specific clients isgranted by Knowledge Technology,

provided that the base fee of $3.00 percopy, plus $0.25 per page is paid

directly to the Copyright ClearanceCenter.

*PC AI (ISSN 0894-0711) ispublished bi-monthly by

Knowledge Technology, Inc.,PO Box 30130,

Phoenix, AZ 85046-0130

EditorialLearning to Communicate

One of the goals of artificial intelligence is to bridge the gap between the world ofcomputers and the world of human beings. Enabling information to pass freely andeasily between those two worlds can lead to many benefits. However simple a conceptthat may seem to be, in practice it is a difficult feat to accomplish. As we all know fromyelling at our computers, they are not terribly good listeners. Also, when our computerstry to communicate with us, the messages can be so complex as to be of little use to theaverage computer user. For example, if you have ever received an error message andclicked on the "show details" button, the resulting details look more like a Scrabblegame that got out of hand than a helpful diagnostic of the problem. Therefore, a largehindrance to information exchange lies in the inability to effectively communicate.There is hope, though. With artificial intelligence technologies that attempt to emulatehuman processes, the wealth of information that can be generated by computersbecomes more accessible to humans.

One such technology that is bridging the information gap is neural networks. Thesenetworks of interconnected processing units mimic the way that neurons in the humanbrain work. Important connections are emphasized while less relevant connections aredowngraded. Neural networks can recognize patterns and predict possible outcomesjust like humans can - but with the advantage of increased speed and capacity forinformation. As I discuss in my article "Neural Nets and Scientific Research: A MatchMade in AI Heaven," this fact makes neural networks an invaluable tool in facilitatingscientific research.

Another technology is intelligent searching tools. Searching through vast amountsof information can be daunting, especially if queries are taken too literally. However,computers, by nature, take everything literally. They use mathematical algorithms toevaluate problems, and are thus governed by the rigidity of mathematics. But there areways to create more efficient and pertinent information searches, as Elizabeth Thededetails in her article "Indexed vs. Unindexed Searching: From Security Classifications toForensics." She presents the differences between indexed and unindexed searching anddiscusses how these enable organizations and individuals to search smarter and faster.

In their article "The Visual Development of Rule-Based Systems," Charles Langleyand Clive Spenser discuss another problem with successful man/machine communication- how information is represented. Most people are more likely to understand a conceptif it is presented visually, whether it be through diagrams, demonstrations, or gestures.In terms of rule-based systems, the information has almost always been presented in atext-based form. The authors contend that rule-based systems will be more effective ifknowledge is presented in a visual form and they discuss the generation of such a system.

In Terry Hengl's article, "Protégé, Ontology and Knowledge Acquisition: KnowledgeRepresentation, the Foundation of Intelligent Systems" he discusses a tool called Protégéwhich is designed to create customized knowledge-based applications. It works on theprinciple of ontologies which are definitions of concepts in terms of a languageunderstandable to all parties involved. Ontologies also delineate relations betweenindividual concepts so as to further define their meanings. This makes sure that everyoneis "on the same page" and that knowledge is fully developed and useful.

As always, PC AI 's regular features are back. Test your knowledge of AI terms withthe AI-Q crossword puzzle, learn about news in the artificial intelligence world with "AIand the Net," find a book or two to read in "The Bookzone," and discover new productswith the "Product Update" and "Buyer's Guide." We hope you enjoy this issue of PC AIand learn something about the many intelligent ways that knowledge is conveyed in theinformation age.

Ilana Marks

PC AI 7 18.3

Product UpdatesProduct Updates

HaleyAuthority 5.0 for .NET and JavaApplicationsHaley Systems, Inc. announces the release ofHaleyAuthority 5.0, a business rules management tool forboth .NET and Java environments. HaleyAuthority allowsbusiness and IT users to develop applications collabora-tively while reducing the steps required to change ruleswhen business conditions change. HaleyAuthority 5.0offers natural language authoring and automatic code gen-eration for .NET and Java applications, along with apatent-pending rules methodology. It offers multiplemodes of rules authoring functionality which includes trueEnglish text, cascading menus, and table formats.HaleyAuthority 5.0 eliminates the need to translate busi -ness requirements from English into complex if-then logicand allows both business and IT users to manage businesspolicies, procedures, regulations, constraints, etc. in anintuitive user interface.

Haley Systems, Inc.info@haley.comwww.haley.com

Business Applications

PC AI 8 18.3

Automated Data Mining Model Developmentand Validation SoftwareNeuralWare announces the release of the NeuralSight data miningmodel development/validation software. NeuralSight augmentsand enhances the NeuralWorks Predict® platform, which enablesresearchers and analysts to create neural network models for pre-diction, classification, and clustering applications. NeuralSightextends the features of Predict by providing a Graphical UserInterface for handling large datasets and by allowing modelers tospecify model performance requirements. NeuralSight selects andranks the best neural networks from a set of networks built byPredict during an unattended model building session. NeuralSightis available now for Microsoft Windows® 2000 and Windows XP.Other operating systems will be supported in 2005.

NeuralWareinfo@neuralware.com

www.neuralware.com

ADVISOR Enterprise 6.1BNH Expert Software announces the release of ADVISOREnterprise v. 6.1. ADVISOR Enterprise is a decision sup-port tool that helps organizations manage training budgetsand resources and identify methods of running programsmore efficiently. ADVISOR contains four modules: Module1 is "Align Training with Organizational Goals". This is aneeds assessment tool for analyzing goals and identifyingwhat is needed to achieve those goals. Module 2 is"Improve Human Performance." This is a decision supporttool that analyzes deficiencies in performance and recom -mends training and solutions. Module 3 is "Select the RightBlend of Delivery Options." This module analyzes theeffectiveness of training courses and then determines the

most cost-effective way to deliver the training. Module 4 is"Manage Training Budgets and Resources." This moduledetermines how much money must be allocated to providethe most benefits. For more information on ADVISOREnterprise, visit the BNH website.

BNH Expert Softwarebahlis@bnhexpertsoft.com

www.bnhexpertsoft.com

PC AI 9 18.3

Data Mining and Modeling

Decision Support

New Version of DocuLex Desktop Search DocuLex and dtSearch announce the release of Version 6.3 ofDocuLex Desktop Search. The resulting product joins DocuLex'sdocument capture technology with dtSearch's text searching tech-nology. Desktop Search works with DocuLex's multiple OCR andother imaging products, including DocuLex's Professional Captureand Office Capture as well as Goby Capture. The integrated prod-uct works to index and search the full text of imaged documents.For example, after DocuLex OCRs a collection of documents

into PDF "image with hidden text" format, Desktop Search canthen index and search that collection, highlighting hits on thescanned image. Desktop Search can also combine a search ofimaged files with other PDF, HTML, XML, email (includingattachments), word processor, database, spreadsheet, presentationand Unicode files. The product displays all retrieved files in abrowser with highlighted hits, while keeping embedded XML,PDF and HTML formatting, links and images intact.

Doculex, Inc.sales@doculex.com

www.doculex.com

dtSearch Corp.sales@dtsearch.com

www.dtsearch.com

MATLAB RF Toolbox 1.0.1The MathWorks announces version 1.0.1 of their RadioFrequency (RF) Toolbox. The toolbox extends the MATLABtechnical computing environment with functions and a graphicaluser interface for working with, analyzing, and visualizing thebehavior of RF components. The toolbox allows the user to spec-ify RF filters, transmission lines, amplifiers, and mixers by theirnetwork parameters and physical properties. The user can alsoread and write industry-standard file formats for network parame-ters. RF Toolbox functions are executed from the MATLAB com-mand line or the RF Tool GUI, or can be called with individualMATLAB scripts and functions. The toolbox includes rectangularand polar plots and Smith® charts for visualizing data.

The MathWorks, Inc.www.mathworks.com

MPI-XF Support for AMD Opteron™ProcessorsEngineered Intelligence (EI) announces the release of MPI-XFversion 1.2, a high performance solution for the industry standardMessage Passing Interface (MPI), and announced validation for

PC AI 10 18.3

Intelligent Tools

Intelligent Search Tools

Indexed Vs. UnindexedSearching: From Security

Classifications to Forensics

Both indexed andunindexed searching have theirplace in the enterprise.Indexed text retrieval istypically more efficient foruses such as generalinformation retrieval andsecurity classification systems.But unindexed searching toohas its place -- in outgoing

Indexed Vs. UnindexedSearching: From Security

Classifications to Forensics

email filtering, searching oflive data sources like RSS newsfeeds, and sometimes inforensics. This article willattempt to explain which searchtechnique to use when, andwhy.

Overview: Indexed TextRetrieval

Indexing the inevitablemillions of documents that anysizeable organization generateson shared file servers is thefastest way to facilitate dataretrieval. An index willtypically store each uniqueword in a document collectionand its location within eachdocument. Indexing also

PC AI 18 18.3

By Elizabeth ThedeBy Elizabeth Thede

works with non--document data,e.g. for forensics searchpurposes (see below).

After indexing, full-textsearch speed, even acrossmillions of documents, istypically less than a second.While indexing a very largecollection of documents for thefirst time may be timeconsuming, subsequent updatesof the index are usually muchfaster. dtSearch, for example,simply checks the filemodification dates of allindexed files, and onlyreindexes those files that havebeen added, deleted or changedsince the last index update.(While the text retrievalterminology here relies on thedtSearch product line, theconcepts in this article aregenerally applicable.)

In addit ion to enabl ingprecis ion boolean searching,

an index can a lso store suchinformation as word positions,enabling word or phraseproximity searching. An indexcan also hold informationabout word frequency anddistribution, enablingcomputation of naturallanguage relevancy rankingsacross a document collection.If the company name appearsin two million documents, itwould get a low relevancyranking. If the latestmarketing terminology appearsin only four documents, itwould get a much higherrelevancy rank. In that way,PR could, for example, enter awhole paragraph of proposedtext for a press release as anatural language search, andzoom right in on the mostrelevant documents.

But full-text searching,whether boolean, natural

language, or otherwise, is onlypart of the text retrievalanswer. Suppose HR wants tolimit its search to documentswith an HR executivedesignation. This type offielded data classification canresult from fields or meta datainside a document, or from anoverlaying documentmanagement-type application.With the latter, fielded dataclassification can rely onassociated database entries,such as SQL or XML, or theaddition of fields "on the fly"during the indexing process.

Adding in SecurityClassifications

Now suppose the goal is toenable searching organization-wide, but to keep the wrongdocuments out of the wrong

PC AI 19 18.3

(Text Continued on pg. 21)

Neural Nets and ScientificDiscovery: A Match Made in

AI Heaven

IntroductionThe human brain is an invaluable

and amazing organ. Electrical andchemical messages bouncing aroundin the brain lead to actions, speech,thoughts, and all other physical andmental properties that define thehuman condition. One of the mostinteresting abilities that the brainaffords us is the ability to recognizethings, people, scents, and otherphysical stimuli. When you thinkabout the diversity even within acertain classification of object, itbecomes more amazing. For instance,within the classification of tree,there is a wide variety of differenttypes of trees, many of which lookdrastically different. A pine treelooks very different from an oak treeand yet we realize that they are bothtrees. We recognize generalcharacteristics of a tree and then areable to expand our definition whenother specimens are encountered. Wealso recognize different states in thelife of a tree - for example, werecognize that leaves may changecolor or fall off. Despite suchseasonal anomalies, a tree is clearlyrecognizable.

In addition to the ability torecognize things, the brain alsoallows us to predict outcomes andoccurrences. When dark cloudsappear in the sky, we can predict thatrain is imminent (unless, of course,

Neural Nets and ScientificDiscovery: A Match Made in

AI Heaven

the weather report says it willrain - in that case, the sky willsuddenly become clear!) Sportsfans try to predict the outcomeof a game based on the previousperformance of the teams.Given a set of circumstances, itis possible to deduce the mostlikely result. Typically, the largerthe set of circumstancesavailable to make a predictionon, the more likely it is to comeup with the correct results.

Recognition and predictionare both highly involved inscientific research. Since scienceis based on other science it isimportant to recognize where aresult is reminiscent of aprevious discovery. Recognizing thoseconnections allows the researcher toexpand their understanding byapplying information garnered fromprevious research. A cornerstone ofgood scientific research is theformulation of an appropriatehypothesis. A hypothesis is nothingmore than a prediction about whatwill occur. Therefore, prediction is abackbone of research. Scientists areconstantly striving to increase theirknowledge base in order to makemore accurate predictions about newexperiments.

Neural Nets and ResearchRecognition and prediction may

be important to good research,however in this day and age, speed isalmost more important. Universitiesdemand that professors publishpapers often and the public demandsnew drugs and technology. The timeneeded for scientists to learn allabout a variety of different fieldsmakes it prohibitive for them tomake truly accurate predictions. Inaddition, a certain type of scientistmay not recognize a certainoccurrence because it may fit more inthe knowledge base of another typeof scientist. With the vast increase inscientific discovery over the lastseveral decades, a mountain ofknowledge is quickly being created

PC AI 24 18.3

By Ilana MarksBy Ilana Marks

Figure 1: The human brain

and its growth shows no signs ofstopping. The human brain simplycannot deal with so much information.Therefore, it is necessary to employartificial intelligence techniques in theprocess of scientific research.Artificial intelligence providesincreased speed and "capacity"compared to the human brain, whichproves incredibly useful toresearchers.

Specifically, neural networks are avery useful tool. Neural networksmimic the behavior of the humanbrain, so they are often used inapplications where systematic thoughtprocesses are important. Processingunits are interconnected andcommunicate like biological neurons.Successful and relevant connectionsare emphasized whereas irrelevantconnections are downgraded. Thismimics the conditioned learning ofbiological organisms. This dynamiclearning is vital in the researchprocess since science does not remainstatic.

Cancer PredictionCancer is a difficult disease to

research. Cancer results from everyday

cells that, in effect, "lose their mind."In normal, healthy cells, the processof cell replication is highly regulated.Many "checkpoints" are in place tomake sure that all steps are doneproperly. These "checkpoints" areusually molecules that are created ordegraded which signal the cell toeither hold in the stage where it is orto move on to the next stage of celldivision. For example, one suchmolecule believed to be involved inseveral different cancers is calledp53. This molecule is often absent ormutated in extracts from tumor cells,indicating that perhaps p53 is amolecule that signals the cell to stopdividing. If p53 is not produced,then cells will divide too rapidly andcancer can result.

Since no two cancer physiologiesare exactly alike, it is notoriouslydifficult to make a prediction aboutwhether the patient is likely tosurvive. Even with two cases of thesame type of cancer, one person maysurvive for many years while anotherwill die within months. A myriad ofdifferent physical factors contributeto the outcome. This is one reasonwhy cancer is such a frightening

disease - no one can tell you whetheryour particular complement of geneswill put you at an advantage ordisadvantage.

Researchers at the NationalCancer Institute (NCI) are working oncreating a model that will help predictthe prognosis of cancer patients.Particularly, they are working with atype of cancer called Neuroblastoma.Neuroblastoma is a childhood cancer.It usually begins with cells of theadrenal gland and spreads, creatingtumors in the neck, chest, abdomen orpelvis. Using DNA microarrayanalysis (see PC AI Volume 18, Issue2 - "Microarrays and ArtificialIntelligence") the researchers studiedthe gene expression profiles of cancerpatients. The microarrays consisted ofabout 25,000 genes and the analysiswas repeated for 49 patients. In orderto connect the microarray results withcertain outcomes, the 49 patients werechosen because their outcomes wereknown. Some of the patients survivedfor more than three years without anycancer-related issues. Others died dueto the disease. Feeding thisinformation into an artificial neuralnetwork, the researchers found they

PC AI 25 18.3

The Visual Development ofRule-Based Systems

The Visual Development ofRule-Based Systems

IntroductionIn the late 1980's Knowledge

Based Systems (KBS) were seento be leading edge softwaretechnology. Developers thoughtthat the simplest KBS paradigm,Expert Systems, perhapscombined with probabilistic andfuzzy logic extensions would soonrevolutionise the way thatsoftware was used throughoutbusiness and other sectors of theeconomy.

KBS software was built onrules which encoded theknowledge of experts in anygiven domain. Computers wouldthen use this encoded knowledgeto make decisions on behalf oftheir human users.

It was not long however,before the bubble of hypesurrounding these systems beganto burst. Something was wrong,but what was it?

The KnowledgeAcquisition Bottleneck

Apart from the limited powerof the computers available at thetime, the major problem was thedifficulty of acquiring implicitknowledge from the minds ofexperts and then representing itexplicitly. This so-calledKnowledge Acquisition Bottleneckwas believed to be the limitingfactor on building systems thatcould do complex, useful tasks.

By the end of the twentiethcentury however, universitydepartments were working hard atthis problem. Curiously it was oftenPsychology departments ratherthan Computer Sciencedepartments which had the mostimpact in this area.

In particular, Ethnography (bythen seen as a core part ofCognitive Psychology) was beingused to study behaviour in situ withthe aim of identifying the cognitiveprocesses underlying thatbehaviour. Just as Margaret Mead

(an early ethnographer) had livedamongst native tribes in PapuaNew Guinea in order to studytheir cognitive behaviour, soPsychology departments weresending researchers (often undercover) into workplaceenvironments to discover howpeople approached problem-solving activities.

This work was, and continuesto be, very successful. Knowledgeacquisition is no longer the 'blackart' it was deemed to be. Despitethis, KBS has continued to beunderused. Why might this be?

A KnowledgeRepresentationBottleneck?

It is my contention that theproblem was not primarily withhow we obtained knowledge, butwith how we represented it. I amnot arguing that rules (orBayesian networks and otherknowledge representationmethods) are inadequate to the

PCAI 29 18.3

By Charles Langley and Clive SpenserBy Charles Langley and Clive Spenser

Order the new Volume 17CD at www.pcai.com/store

task, but rather that it is the way in which these rulesand other representational formalisms are themselvesrepresented that is the limiting factor.

At first a simple rule-base is relatively transparent,especially if properly documented. Certainly suchsystems were easier to comprehend than proceduralcode and were subsequently easier to update andamend. As such rule bases became larger and morecomplex however, a simple syntax error, perhaps onlyinvolving one word, could prevent them fromoperating correctly. The complexity of these rulesetsalso meant that it was difficult to get an overview ofwhat was intended, thus impeding their maintenanceand extension.

A Picture is Worth a Thousand WordsThe problem of rulebase comprehensibility, I

would argue, is the fact that we have primarilyrepresented knowledge using text based structuresrather than visual ones. No matter how close tonatural language a knowledge representation languageis, you cannot see at a glance what a complex system istrying to do.

Visual Rule GenerationRule generation via a graphical interface is a hot

topic right now, with offerings from a number ofcompanies small and large. This is being driven in partby current interest in so-called 'business rulesmanagement' which is arguably a reawakening of theKBS paradigm we mentioned earlier.

London based Logic Programming Associates is anappropriate company to enter this market as it has beenproducing rule-based software since the mid 1980s. Its latestproduct, VisiRule, enables rule-based systems to beautomatically generated from a flowchart drawn on thescreen.

Consider the following business rule (Ross 2003):

Rule: An order must be credit-checked if any of thefollowing is true:

* The order total is more than $500

* The outstanding balance of the customer's account plus the order amount is more than $600

* The customer's account is not older than 30 days

* The customer's account is inactive

* The customer is out ofstate

PCAI 30 18.3

Figure 1

Protégé, Ontology andKnowledge Acquisition:

Knowledge Representation,The Foundation of Intelligent

Systems

Protégé, Ontology andKnowledge Acquisition:

Knowledge Representation,The Foundation of Intelligent

Systems

IntroductionKnowledge is a decisive competitive-

advantage for today's corporations.Knowledge of schedules, raw materials,labor, manufacturing and distribution isessential to the supply chain whileknowledge of customer interests andbuying habits, latest technologies, budgetconstraints, marketing plans are crucial toproduct development. It offers a powerfultool for gaining market share andpreserving a competitive edge, but it iscostly to capture and control.Methodologies and technologies thatassist in the acquisition, maintenance,and distribution of knowledge areessential to an organization's success.Today's society, and the world in general,have contributed to this growth inimportance of knowledge managementand knowledge based systems. Someexamples include (www.worldedreform.com/intercon2/f15a.pdf):

* Accelerating rate of change in every aspect of technology and society

* Staff migration and attrition (downsizing and reengineering)

* Geographic dispersion associated with globalization of markets

* Global integration of cultures, companies and markets

* Increase in networked organizations* Increased level of education and

training of the population* Growing knowledge-intensity of

goods and services* Revolution in information technology

It is not easy to efficiently and cost-effectively identify, acquire and maintainthis knowledge. At a minimum,organizations must be able to:

* agree to an organization-wide vocabulary to ensure knowledge is consistently communicated and understood;

* identify, explicitly represent and modelthis knowledge;

* share and reuse this knowledge across independent applications and domains.

This article looks at these aspects ofknowledge acquisition by examiningProtégé (http://Protégé.stanford.edu), a freeopen-source Java tool with an extensiblearchitecture for creating customizedknowledge-based applications - based onOntologies (www.ontology.org). It alsoreviews the concept of Ontologies (see sidebar), and associated support methodologies,which establish the vocabulary and modelthe concepts along with their inter-relationships. This concept also includesprocessing of the associated attributes for aparticular field of knowledge. By reviewingthe evolution of Protégé, an ontologymodeling and knowledge acquisition

environment (developed by StanfordMedical Informatics(http://camis.stanford.edu) at the StanfordUniversity School of Medicine), anunderstanding of fundamental conceptsthat underpin knowledge acquisitionemerges. This environment creates andmodifies the ontologies and knowledgebases it generates to enable developers anddomain experts to build knowledge-basedsystems.

Other associated technologies andstandards mentioned in this articleinclude the Open Knowledge BaseConnectivity (OKBC), Generic FrameProtocol (GFP), Resource DescriptionFramework (RDF), OWL and OWL-S.

The Evolution of ProtégéOpal (http://smi-web.stanford.edu/

pubs/SMI_Abstracts/SMI-86-0137.html)was an expert system shell-basedapplication developed as part of themedical domain Oncocin(http://citeseer.ist.psu.edu/context/1419258/0). Oncocin developed this knowledgeacquisition and advice system forprotocol-based cancer therapy. Opalenabled patient history entry by the

PC AI 37 18.3

By Terry HenglBy Terry Hengl

Order the new Volume 17

CD at www.pcai.com/store

physician or nurse (the domain experts),resulting in a suggested treatment or test.The knowledge base, a collection of if-then rules and other data, captured theclinical protocols. Opal translated theexpert's input, via graphical forms, intoan internal representation specificallytailored for Oncocin. This projectidentified three different levels ofknowledge for this particularinformation:

1. Structural domain concepts used by the knowledge engineer to create the Opal knowledge-acquisition application;

2. The domain expert (oncologist) knowledge - oncology protocols;

3. Case data entered by the user to exercise the expert system decision capability.

Since Opal was inference engine based,it enabled reuse by knowledge engineersto create different knowledge bases -ultimately creating domain specific expertsystems. The knowledge engineer wasresponsible for the knowledgeacquisition; a tedious and time-intensivetask. Unfortunately, the concept of aknowledge engineer separated thedomain experts from the domainknowledge bases and this separationintroduced a potentially large source ofincorrect knowledge.

In 1987, Mark Musen built an

application forknowledge-basedsystems with a goalof buildingknowledge-acquisitionapplication as part ofOncocin. Based onthese three differentlevels of knowledge,he believed thatknowledgeacquisition occurs inphases withknowledge obtainedin one phasedefining thestructural knowledgefor subsequentphases. Musen's goalwas to reduce thework engineers didto constructknowledge basesduring knowledge-acquisition. Henoticed thatknowledge obtainedduring a specificphase influences theknowledge relatedapplication requiredfor later stages.

From this earlyconcept, Protégéevolved through four

phases, resulting in a rich developmentenvironment available for both researchand knowledge-management. (seehttp://smi-web.stanford.edu/pubs/SMI_Abstracts/SMI-2002-0943.html for moreinformation on the evolution of Protégé.

Protégé-I - Knowledge BasesThis early phase, where Protégé-I

simplified the knowledge acquisitionprocess for building medical expertsystems, learned from the earlier OPAL-based system. The intent of the tool was tosimplify the knowledge acquisition processfor the knowledge engineer, alreadyoverloaded with complex tasks to perform- a key issue with early expert systemdevelopment. One goal of Protégé was toprovide an application that createdKnowledge Acquisition tools (KA) from aformally defined collection of concepts.This enabled the domain expert to createthe knowledge base, eliminating the timeconsuming process of the knowledgeengineer learning the domain. Earlyassumptions for Protégé 1 included:

PC AI 38 18.3

Multiple PSM – Domain-IndependentPSM and automated form generationDefined three classes of ontology:• Domain, Method and Application

1986 1988 1990 1992 1994 1996 1998 2000 2002

Opal

Protégé-I

Protégé-II

Protégé/Win

Protégé-2000

Pre-Protégé

Single PSM - Episodic Skeletal-Plan Refinement (ESPR)PSM defines semanticsKnowledge bases are problem-specificBased on Xerox LISP Machine

Integrated Tool SetBased on MS Windows

Includes OKBC Knowledge ModelRuns under Java VM

Translated expert’s case entry into Oncocin’s internal representation• Structural domain concepts,• oncology protocols• Case data exercise the expert system decision capability

Figure 1: Evolution and enhancements of Protégé from 1986 to present.

PC AI 49 18.3

Know Your AI-QKnow Your AI-QWelcome to PC AI’s feature that allows you to test your AI-Q. This is the next in a series

of crossword puzzles on the various technical categories of AI. Future puzzle topics willinclude robotics, LISP, AI languages, expert systems, agents and many more. Theanswers will appear in the next issue.

This issue’s subject is Neural Networks, Fuzzy Logic, and Speech Recognition.

PC AI 51 18.3

Solution to 18.2 AI-QCrossword Puzzle

The topic of this puzzle was Machine Learning

Virtual Museum EducationAssistant

The Science and TechnologyMuseum in India has integrated a virtualeducation assistant called a "Cyberlady"that helps visitors learn about the exhibitsand answers questions. "Cyberlady" wasdeveloped through a collaborationbetween the museum and the Centre forDevelopment of Imaging Technology.When a user types in a question, the"Cyberlady" uses artificial intelligencetechnology to formulate a response thatgrabs the user's attention and providesinteresting information. The "Cyberlady"responds using a synthesized voice. The"Cyberlady" learns from theconversations and stores the informationgleaned from interacting with visitors sothat new answers can be developed.

In addition to the "Cyberlady," themuseum has also integrated virtuallaboratory modules that simulate variousscientific experiments. One modulesimulates the flame test whereby a rod isdipped in various chemicals and then isexposed to flame. The flame changescolor based on what type of chemical ison the rod. Another module simulates thefour-stroke engine. In this module across-section of an engine is observedand all of the components are identified.Other modules are in the works for thefuture.http://www.hindu.com/lf/2004/09/24/stories/2004092400500200.htm

AI Learns How to CramWe all remember those late night

study sessions from college where wedesperately tried to fill our minds with asemester's worth of information in a fewhours. Well, now there is a tool poweredby AI that helpsstudents to studysmarter. The tool, calledCram101, is available atmany universitybookstores. Cram101distills information fromtextbooks and organizesit into outline form. Theoutlines are printable sothat students can takethem to class andaugment them withinformation from theprofessor's lecture.

Cram101 alsocreates practice teststhat give students ameans of diagnosing thelevel to which they havelearned the material.Since students oftenhave short attentionspans and bore easily,the practice tests arepresented in the form ofa game. The tests aredesigned to be a self-teaching tool rather thanan accuraterepresentation of the

types of tests found in the classroom.Therefore, the answers to questions giveclues to the answers of other questions.With those connections, the student has abetter chance of remembering theinformation. Cram101 membership is

PC AI 52 18.3

A I and the NetA I and the NetVirtual MuseumVirtual Museum

EducationEducationAssistant,Assistant,

Call CenterCall CenterComputers,Computers,

Nose Mouse,Nose Mouse,Network ofNetwork of

RoboticRoboticTelescopes,Telescopes,

and moreand more

Ilana MarksIlana Marks

preferentially in the future for support. Asa result of these connections, a leadershipstructure emerges. To optimize themodel, the researchers are adjusting theconnectivity of this leadership so that theinfluence of those agents is regulated.The information gleaned from this modelcould lead to the development of multi-agent systems suited to applicationswhere human intervention has beenrequired in the past. Since the agentslearn from each other, the humancomponent becomes less important.www.trnmag.com/Stories/2004/092204/Agent_model_yields_leadership_092204.html

Network of RoboticTelescopes

With many astronomical eventsoccurring in a very short span of time, itis important to be able to react quickly inorder to observe them. Anyone who haswatched a meteor shower knows that thisis true. If you are not looking in the rightplace at the right time you will miss theshow. However, if there are many peoplewatching the meteor shower from manydifferent positions then it is possible to

observe more of the action.British astronomers are

hoping to take advantage of theinformation-capturing power ofrobotic telescopes by linkingthem together and controllingthem with artificially intelligentsoftware. The network is calledRoboNet-1.0. Since the robotictelescopes in the network arelocated around the world, theconnection provides a widerrange of view as well as theability to react quickly tointeresting phenomena.Researchers hope to use thisnetwork to study gamma raybursts. Gamma ray bursts arevery intense energy bursts. Thebursts are detected frequently;however the longest ones last

Model of a Multi-AgentSystem

In a joint research project from LosAlamos National Laboratory, theUniversity of Houston and RensselaerPolytechnic Institute, a model of anetwork of agents has been created. Thegoal is to use the model to predict thebehavior of multi-agent systems - a taskthat can be very difficult. The model isbased on the "minority game." In thisgame, agents make decisions and try to bein the minority when the results arerevealed. The agents learn fromexperience what strategies are successful.In addition, the researchers created socialnetworks amongst the agents so thatinformation about strategy could beshared. Therefore, when makingdecisions a single agent will consider itsown experience as well as the experiencesof its neighbors. The agents strengthenconnections with agents that provideduseful recommendations in the past. Thiscan be compared to human socialinteraction - we come to value the adviceof certain people and thus turn to them

available for a monthly fee.www.fsunews.com/vnews/display.v/ART/2004/09/23/4151f0e8c872a .

Call Center ComputersA new AI tool aims to take the

frustration out of call center assistance.Once a person finally gets through to acustomer service representative, oftenthey will have to wait additional timewhile the operator searches through thecomputer to find the answer they arelooking for. IBM is developing acombination speech recognitionutility/search engine that listens in to thesupport phone calls and begins searchingfor the required information before thecaller has finished their request. The toolworks by using speech recognition topick out key words indicating the caller'strouble. Those keywords are then enteredinto the call centre database, giving theoperator a head start on pulling up theinformation.

In addition to assisting the operatorin finding the correct information, thesystem can also alert the operator toimportant points that must be stressed -especially if those points constitute legalwarnings. Just as the system listens to thecaller's end of the conversation, it canalso listen to the operator and provideon-screen reminders in order to ensurethat the operator handles the callproperly. While the software is still in itsinfancy with only a few phrases andwords identified, commercial versions arein the works with a trial versionscheduled to be implemented in a Dutchbank.www.newscientist.com/news/news.jsp?id=ns99996430 .

PC AI 53 18.3

Order Volume 16on CD

6 Issues plus abonus issue

Visit the PC AIStore

Volume 17Is Now Available

Applying UML by RobPooley and Pauline WilcoxThis book addresses issues faced byusers in adopting the UnifiedModeling Languages (UML) andhelps them to apply it. The bookcovers UML in depth, includingnotation on profiles and extensions.

The book assumes prior experience insoftware engineering or businessmodeling, an understanding ofobject-oriented concepts and a basicknowledge of UML.

Table of Contents:Preface1. Introduction2. A Complete Example3. Issues and Features4. Graphics and Interaction Based

Applications5. Business Model6. Embedded Control7. Reuse8. Review of Case Studies in

Chapters 4,5,6, and 79. The Need for Methodologies10. The Capability Model11. Evaluation of MethodologiesAppendix A - UML NotationAppendix B - UML SemanticsAppendix C - Code Generation andRound Trip EngineeringReferencesIndex

Applying UML by Rob Pooley and

Pauline Wilcox, November 2004,

Morgan Kaufmann Publishers,ISBN: 0-7506-5683-2, Pages: 224.

For more info:http://books.elsevier.com/us/mk/us/subindex.asp?isbn=0750656832&country=United+States&community=mk&ref=&mscssid=WN7TTGGDMB8T8LG87CEADJGF0MGDC5R7

Text Mining: PredictiveMethods for AnalyzingUnstructured Information bySholom M. Weiss, Nitin Indurkhya,Tong Zhang and Fred DamerauText mining allows users to find trendsand patterns in text-based information.This book analyzes new and proventechniques in text mining. It discussestopics such as automated documentindexing and information retrieval andsearch. Also included is a look at newresearch in text mining includinginformation extraction and documentsummarization.

Table of Contents:Preface1. Overview of Text Mining2. From Textual Information to

Numerical Vectors3. Using Text for Prediction4. Information Retrieval and Text

Mining

PC AI 58 18.3

By Ilana MarksBy Ilana Marks

Applying UML

Text Mining: Predictive Methods forAnalyzing Unstructured Information

Spoken Dialogue Technology

Fuzzy Control of Queuing Systems

Data Modeling Essentials, Third Edition

Requirements Engineering

The BookThe BookZoneZone

5. Finding Structure in a Document Collection

6. Looking for Information in Documents

7. Case Studies8. Emerging DirectionsAppendix: Software NotesReferencesAuthor IndexSubject Index

Text Mining: Predictive Methods

for Analyzing UnstructuredInformation by Sholom M.

Weiss, Nitin Indurkhya,

Tong Zhang and FredDamerau, 2004, Springer

Verlag, ISBN: 0-387-95433-3,

Pages: 236.

For more info:http://www.springeronline.com/sgw/cda/frontpage/0,11855,4-146-22-

34526885-0,00.html

Spoken Dialogue Technology: Towards theConversational User Interface by Michael F.McTearThis book covers spoken dialogue systems, ranging fromtheoretical aspects to a detailed analysis of well-establishedmethods and tools for developing spoken dialogue systems. Thebook enables the reader todesign and test dialoguesystems. Developmentenvironments and languagesinclude the CSLU toolkit,VoiceXML, SALT, andXHTML+ voice. Research inspoken dialogue systems ispresented along withtheoretical issues. A dedicatedweb site containingsupplementary materials, codeand links to resources isavailable to readers.

Table of Contents:1. What is a Spoken

Dialogue System?2. The Components of a

Spoken Dialogue System3. Describing Dialogue4. Developing a Spoken Dialogue System: The Dialogue

Engineering Lifecycle5. Directed Dialogue Systems6. Developing Directed Dialogue Systems using the CSLU

Toolkit7. Developing Directed Dialogue Systems using VoiceXML8. Mixed-Initiative Dialogue Systems9. Developing Mixed-Initiative Dialogue Systems using

VoiceXML10. Developing Mixed-Initiative Dialogue Systems using the CU

Communicator System11. Conversational Agents12. Future Developments

Spoken Dialogue Technology: Towards the Conversational User Interface

by Michael F. McTear, 2004, Springer Verlag, ISBN: 1-85233-672-2, Pages: 432.

For more info:http://www.springeronline.com/sgw/cda/frontpage/0,11855,4-40109-

22-28205661-0,00.html

Fuzzy Control of Queuing Systems byRuntong Zhang, Yannis A. Phillis and Vassilis S.KouikoglouQueuing controlaffects manufacturingand communicationnetworks around theworld. This bookdiscusses the use offuzzy logictechnologies to solvequeuing controlproblems. Thisapproach determinesexplicit solutions tovarious types ofcontrol issues inqueuing systems.Included in the bookare detailed casestudies to demonstratethis new approach andhow it differs from classical techniques. The book is directed atstudents, practitioners and researchers.

Table of Contents:Preface1.Introduction2.Fuzzy Logic3.Knowledge and Fuzzy Control4.Control of the Service Activities5.Control of the Queue Discipline6.Control of the Admission of Customers7.Coordinating Multiple Control Policies8.Applications of Fuzzy Queuing Control to the Internet9.ClosureAppendix: Markov Queuing Models and SimulationReferencesIndex

Fuzzy Control of Queuing Systems by Runtong Zhang, Yannis A.

Phillis and Vassilis S. Kouikoglou, 2004, Springer Verlag,

ISBN: 1-85233-824-5, Pages 175.

For more info:http://www.springeronline.com/sgw/cda/frontpage/0,11855,4-40109-

22-29559980-0,00.html

PC AI 59 18.3

Visit the PC AI store:

www.pcai.com/store

The vendor’s address, phone number, e-mails, andURL are in a separate table after the productinformation table.

Next issue: Knowledge Representation &Management, Expert Systems, ExpertSystem Development Tools, MachineLearning, Logic and Reasoning Systems,Intelligent TutorsContact PC AI for a submission form at bg@pcai.com

Buyer’s GuideFuzzy Logic, Fuzzy SQL, Intelligent Process Control,

Neural Networks, Speech Recognition,Text Mining, Web Utilities

Special Note on System Requirements: When MS Windows appears, this assumes Win 95. 98, NT, ME, 2000,XP, A Pentium Processor, 32MB RAM and 32MB Disk space available unless stated otherwise.

Fuzzy LogicProduct

Company Description SystemRequirements Price

FLINTLPA

Fuzzy Logic toolkit with support for various treatments of uncertainty includingBayesian Updating and Certainty Factors. Includes Fuzzy Editor and support for fuzzyrule matices.

MS Windows ContactVendor

Fuzzy LogicToolbox Version2.2

TheMathWorks, Inc.

Extends MATLAB computing environment. Contains tools for the design of fuzzylogic based systems. Supports many common fuzzy logic methods including fuzzyclustering and adaptive neurofuzzy learning. Includes special GUIs, membershipfunctions, support for AND, OR, NOT logic within rules, standard Mamdani andSugeno-type fuzzy inference systems, automated membership function shaping. Fuzzyinference systems designed with the toolbox can be embedded in a Simulink model andC code or stand-alone executable fuzzy inference engines can be generated.

Contact Vendor ContactVendor

fuzzyTECH OnlineEdition

Inform SoftwareCorporation

Includes all features of the fuzzyTECH Professional Edition. Generates systems whichcan be remotely optimized on-the-fly. Fuzzy logic system is designed, tested, andgenerated as C code on the PC. Then, the C code is implemented and started on thetarget hardware. Communicates bi-directionally with the implemented code on thetarget system at any time via a serial cable, a network file system, or anothercommunication link. Includes the NeuroFuzzy Module and a distribution license forthe fuzzyTECH Viewer, a monitoring-only version of fuzzyTECH that can bedistributed with every runtime system generated by fuzzyTECH.

Contact Vendor ContactVendor

fuzzyTECHProfessionalEdition

Inform SoftwareCorporation

Universal fuzzy logic design system for all technical application areas. Generatesportable C code that can be flexibly adapted to any target hardware and deploys fuzzylogic runtime systems as DLL/ActiveX modules. fuzzyTECH's code allows for use inreal-time control systems. Supports advanced fuzzy logic inference methods(compensatory operators, fuzzy associative maps, s-shape membership functions,arbitrary membership functions, etc.), and is also compatible for complex fuzzy logicapplications. Allows Online debugging with the PC-based runtime modules.

Contact Vendor ContactVendor

Mathematica FuzzyLogic Version 2

WolframResearch, Inc.

Set of tools for creating, modifying and visualizing fuzzy sets and systems. Providesexamples to introduce users to fuzzy logic and to show how concepts can be applied.Includes a variety of membership functions, compositions and inferencing, standardand parameterized fuzzy aggregators, fuzzy operators, fuzzy logic control, fuzzyarithmetic and more.

Mathematica 5 orlater, MSWindows, MacOS X, Linux, orUnix

ContactVendor

PC AI 62 18.3

Fuzzy LogicProduct

Company Description SystemRequirements Price

Visual Rule StudioRule Machines

Corp

Microsoft Visual Basic ActiveX Add-on that adds a rule designer and inference engineto Visual Basic. Production Rule Language (PRL) includes fuzzy relational operatorsallowing developers to create rich fuzzy logic applications. Rules are authored within arules editor which is seamlessly incorporated into the Visual Basic developmentenvironment. Rules can be organized by rule sets, and are isolated from Visual Basiccode, making the rule sets easier to build, verify, and maintain. Two sample fuzzy logicprojects are included with the software.

MS Windows, VB5.0 or 6.0

ContactVendor

PC AI 63 18.3

Fuzzy SQLProduct

Company Description SystemRequirements Price

dtSearch TextRetrieval Engine

dtSearch Corp.

Lets developers add dtSearch’s text retrieval to applications. Over two dozen indexed,unindexed and fielded data search options. Hit-highlighted file displays (for PDF andHTML, with images and links). Built-in HTML converters for "Office," XML, ZIPand other popular file types. Supports SQL, XML, .NET, and much more. See websitefor downloadable evaluations.

2 versions:dtSearch Enginefor Win & .NETand dtSearchEngine for Linux.

ContactVendor

Fuzzy Query Sonalysts, Inc.

A database query and data analysis tool that uses fuzzy logic to better represent theunderlying semantics of data analysis questions, and ODBC to provide connectivity toa wide range of existing data stores.

Windows95/98/NT/2000

$59

Intelligent Process ControlProduct

CompanyDescription System

RequirementsPrice

Advisor Enterprise BNH Expert

Software Inc.

Decision support tool manages training budgets and resources. Provides analysis foreffectiveness, cost and impact of each training activity. Data resides in a centraldatabase. Managers can determine required resources to run one or multiple trainingprograms, map where the money is spent (salaries, travel, etc.), identify what workedand why, detect problem areas, assess the impact of alternate delivery options andpotential risks, evaluate build versus buy decisions, and consider multiple "what if"scenarios.

Java script browser(IE 5.5 orNetscape 6.0) fornavigation bar. 32MB of RAM, 100MB of disk space.

$295 for singleuser 1 yearsubscription

Amzi! Prolog +Logic Server

Amzi! inc

Enables the integration of intelligent components with conventional applications forthe addition of business rule logicbases for process control, pricing, configuration,workflow, planning, and problem solving. Access a logicbase of rules like a database.The rules are expressed in Prolog. Encapsulated as a Java Class (JSP and Servlets),C/C++ Class, .NET Class (VB, C#), Delphi Component, and DLL/SO API. Addcustom Prolog predicates in Java, C/C++, C#, VB or Delphi. New Eclipse IDE withsource/remote debugger.

Contact Vendor $0-$1499

CommonKnowledge

ObjectConnectionsAustralia

Cross-platform system for automation and maintenance of business rules throughoutthe application lifecycle. Comprising two parts, a rules engine and a graphical rulesdesign studio, the system provides developers with a means to design, document,implement, and maintain their applications' business rules and logic externally in theform of Decision Tables, Trees, and scripting languages including JS, VBS, Perl andPython. Rule-sets can be changed on-the-fly without recompiling the applications, andare language and platform-independent, ideal for process control and operatorguidance software development. Contact vendor for pricing on the ProfessionalEdition.

MS Windows,Linux/Unix/Solaris,. 128 MB RAM,25-50 MB disk.Microsoft IE V4+.Supported: VB,Delphi, MS .Net,WSH, Java etc.

ContactVendor

PC AI Volume 17 now on CD - 6 issues in both HTML and PDFhttp://www.pcai.com/store

Paid Subscribers $14/CD, Non-Paid Subscribers $32/CD US Postage $3/CD, Foreign $5/CD

PC AI 73 18.3

Web UtilitiesAmzi! Prolog + LogicServer

Amzi! inc

Enables the integration of intelligent components with conventional applicationsfor the addition of business rule logicbases for process control, pricing,configuration, workflow, planning, and problem solving. Access a logicbase ofrules like a database. The rules are expressed in Prolog. Encapsulated as a JavaClass (JSP and Servlets), C/C++ Class, .NET Class (VB, C#), DelphiComponent, and DLL/SO API. Add custom Prolog predicates in Java, C/C++,C#, VB or Delphi. New Eclipse IDE with source/remote debugger.

Contact Vendor $0-$1499

Ask.Me OnLineKnowledge

Broker, Inc.

Available 24-hours a day the Knowledge Library contains 50,000+ solutions toFAQ's about desktop applications, operating systems and the Internet. Searchengine delivers instant answers. Reduce inbound call volume and add solutions toproprietary applications. Use as a resource for a help desk. If an answer is notfound, go straight to eHelpMail. For a small fee, our technicians will research yourproblem and e-mail the answer within 24-hours. Available to individuals,businesses and OEMs.

Access to theInternet

Contact Vendor

dtSearch Web withSpider

dtSearch Corp

Publishes content to a Web site for searching. Over a dozen indexed and fieldeddata search options. Highlights hits in HTML and PDF, while displaying links andimages. Converts "Office," XML, ZIP, etc. files to HTML with highlighted hits.Built-in Spider can also expand searchable database to content on a third-partysite. See website for evaluations, press reviews and case studies.

Runs on IIS-basedWeb sites. Linuxversion of dtSearchText RetrievalEngine is alsoavailable.

Contact Vendor

Visual Rule StudioRule Machines

Corporation

Expert system development tool based on the Production Rule Language (PRL)and Inference Engines of LEVEL5 Object adds a rule designer and inferenceengine to Microsoft Visual Basic. ActiveX add-on allows developers to consolidaterules in rule sets in isolation of presentation and data connectivity code.Production Rule Language (PRL) used to define the rules contends with logicprocessing to write, verify, and maintain rules. Customized web based expertsystems can be developed from within the Microsoft Visual Studio IntegratedDevelopment Environment (IDE). Rule sets can be reused for other web basedapplications as well as Microsoft Windows based systems. Inference engine canuse backward chaining, forward chaining, or a combination of both.

MS Windows, VB5.0 or 6.0

Contact Vendor

Vendor Address

Company Address Phone and Email Web Address

U.S. VendorsAmzi! inc. 47 Redwood Road

Asheville, NC 28804828.350.0350info@amzi.com

www.amzi.com

Alyuda Research, Inc. 1450 Frontero Ave.Los Altos, CA 94024

425.928.3570sales@alyuda.com

www.alyuda.com

dtSearch Corp. 6852 Tulip Hill TerraceBethesda, MD 20816

800.IT.FINDSsales@dtsearch.com

www.dtsearch.com

Gecko Systems, Inc. 1640B Highway 212, SWConyers, GA 30094

678.413.9236info@geckosystems.com

www.geckosystems.com

Inform Software Corporation 222 S Riverside Plz Ste 1410Chicago, IL 60606

312.575.0578ioffice@informusa.com

www.fuzzytech.com

ISYS Search Software 8775 E. Orchard Rd. #811Englewood, CO 80111

800.992.4797info@isysusa.com

www.isysusa.com

KnowledgeBroker, Inc. PO Box 17097Reno, NV 89521

626.441.0702sales@kbi.com

www.kbi.com

Megaputer Intelligence, Inc. 120W, 7th Street, Suite 310Bloomington, IN 47404

812.330.0110sales@megaputer.com

www.megaputer.com

National Instruments 11500 N. Mopac ExpwyAustin, TX 78759-3504

512.683.0100 www.ni.com

PC AI 74 18.3

Vendor AddressCompany Address Phone and Email Web Address

U.S. VendorsNeuroDimension, Inc. 1800 N. Main Street, Suite D4

Gainesville, FL 32609352-377-5144, 800-ND-IDEASinfo@nd.com

www.nd.com

NeuralWare 230 East Main Street, Suite 200Carnegie, PA 15106-2700

412.278.6288sales@neuralware.com

www.neuralware.com

Promised LandTechnologies

195 Church Street 11th FloorNew Haven, CT 06510

203.562.7335support@promland.com

www.promland.com

Rule MachinesCorporation

51745 396th Ave.Frazee, MN 56544

218.334.3960info@rulemachines.com

www.rulemachines.com

SAS Institute Inc. 100 SAS Campus DriveCary, NC 27513-2414

919.677.8000 www.sas.com/index.html

Sensory, Inc. 1991 Russell AvenueSanta Clara, CA 95084-2035

408.327.9000sensorysales@sensoryinc.com

www.sensoryinc.com

Sonalysts, Inc. 215 Parkway NorthWaterford, CT 06385

860.326.3772fuzzyquery@sonalysts.com

www.sonalysts.com

Sprex 1210 NE 124th StSeattle, WA 98125

206.367.7741info@sprex.com

http://sprex.com/index.php

SPSS Inc. 233 S. Wacker Drive, 11th FloorChicago, IL 60606-6307

312.651.3000sales@spss.com

www.spss.com

StatSoft, Inc. 2300 E. 14th St.Tulsa, OK 74104

918.749.1119info@statsoft.com

www.statsoft.com

Text AnalysisInternational, Inc.

1669-2 Hollenbeck Ave. # 501Sunnyvale, CA 94087

408.746.9932 or 877.235.6259info@textanalysis.com

www.textanalysis.com

The MathWorks, Inc. 3 Apple Hill DriveNatick, MA 01760-2098

508.647.7000 www.mathworks.com

The Modeling Agency PO Box 7541The Woodlands, TX 77387

936.321.2177info@the-modeling-agency.com

www.the-modeling-agency.com

Ward Systems Group,Inc.

5 Hillcrest DriveFrederick, MD 21703

301.662.7950sales@wardsystems.com

www.wardsystems.com

Wolfram Research, Inc. 100 Trade Center DriveChampaign, IL 61820-7237

217.398.0700info@wolfram.com

www.wolfram.com

Canadian VendorsBNH Expert Sofware Inc. 4000 Steinberg Street

St. Laurent, QC, H4R 2G7514.745.4010info@bnhexpertsoft.com

www.bnhexpertsoft.com

Universal DynamicsTechnologies, Inc.

100-13700 International PlaceRichmond, BC, V6V 2X8

604.214.3456info@brainwave.com

www.brainwave.com

Vendors Outside North AmericaKDiscovery.com 31 Simei Street 4 #02-20

Republic of Singapore 529877(65) 3442357info@kDiscovery.com

www.kdiscovery.com

LPA Studio 4, RVPB, Trinity RdLondon England, SW18 3SX

+44.20887.12016info@lpa.co.uk

www.lpa.co.uk

MacSpeech, Inc. 81 Buckland StreetChippendale N.S.W. 2008

(02) 9319 2288info@macsense.com.au

www.macspeech.com;www.macsense.com.au

Magenta Technology 33 Glasshouse StreetLondon W1B 5DG UK

44(0)2077436450contact@magenta-technology.com

www.magenta-technology.com

Object Connections PO Box 562Wallsend, Australia, 2287

61249500577info@objectconnections.com

www.objectconnections.com

TEMIS SA 193-197 rue de Bercy75582 Paris Cedex, France

+33(0)140044670info@temis-group.com

www.temis-group.com

Visit PC AI at www.pcai.com

PC AI 50 16.4 200252 P C A I

ADVERTISER INDEX

Select the company name toautomatically transfer to thecompany’s home site of interest

Company Pages

AI Developers 9, 13,

38 ,42,

61

Amzi 10, 21,

28, 66

Applied Business Intelligence 75

dtSearch 8, 11,

19, 54

Exsys Inc 16, 25,

34, 55,

64

Franz Inc 4, 14, 30,

76

Intellicrafters 68

LPA 15, 26,

31, 52

Out of Toner 75

PC AI Back Issues 3

PC AI CD 6

PDC 12, 33,

57, 61

QMC 2

Statsoft 17, 56,

60

The Modeling Agency 75

WizSoft 75

PC AI 75 18.3

Submit your product(s)for the

PC AI Buyer’s Guide to bg@pcai.com

480-922-8922

250 different toner cartridges

hp laserjet 4000/4050 (C4127X) - $79.00

hp laserjet 4200 (Q1338A) - $99.00

hp laserjet 4300 (Q1339A) - $105.00

Call480-922-8922

www.OutOfToner.com

Applied Business IntelligenceApply advanced technology to improve:

Strategic Advantage, Fraud Detection, Identify Theft, Terrorism Risk Analysis, Information Technology Risk Analysis

Available technologies include:

Data Mining Data ArchitectureData Warehousing Business ForecastingBusiness Modeling Business AutomationMachinte Learning Decision SystemsStatistical Analysis Web-based DistributionDeclarative Methodology Neural Networks

Contact us directly at (602) 363-0702

PC AI 76 18.3