Date: September 24, 1999

Speaker: Jeff Heflin (with an introduction by Richard Fikes that will include an overview of XML
                and DTDs)

Title: "SHOE, a Markup Language For Representing Knowledge on the Web"

Abstract:

We describe SHOE, a set of Simple HTML Ontological Extensions that allows Web authors to annotate their pages with machine-readable knowledge that can then be used by agents and query engines.  There are two types of SHOE enabled pages: those that define ontologies and those that declare instances.  SHOE ontologies are made available to document authors and SHOE agents by placing them on the Web.  Each ontology typically extends the base SHOE ontology and may extend other preexisting ontologies.  Instances are distinct entities that can be classified in categories and can have relationships to other instances.  Every instance states which ontologies it is using to make claims.  We also describe an initial set of applications to demonstrate SHOE’s promise, including Expose, a web-crawler that looks for SHOE pages and loads them into a Parka knowledge base; and Knowledge Annotator, a graphical application for annotating Web pages with SHOE.

Date: September 29, 1999

Speaker: Jeff Heflin

Title: "SHOE, a Markup Language For Representing Knowledge on the Web" (continued)

Abstract:

Whereas the previous seminar described the SHOE language, this seminar will focus on applications of the language.  We will demonstrate a number of SHOE tools, including the Knowledge Annotator, a tool for marking up web pages with semantic knowledge, and SHOE Search, a tool for querying this knowledge within the context of an ontology.  We will examine some existing SHOE ontologies and will discuss the on-going project with the U.S. Food and Drug Administration (FDA) to apply SHOE to the food safety domain.  Finally, we will discuss future areas of research and challenges that must be addressed by Web KR languages.  Participants are encouraged to bring issues for discussion.

Date: October 6, 1999

Speaker: Tom Gruber

Title: Creating an Organizational Memory from Knowledge Work

Abstract:

For many organizations, there is nothing more valuable or irreplaceable than the collective knowledge of employees.  Knowledge Management technology helps to capture, share, disseminate and apply this knowledge to practice.  One approach to KM is to create an *organizational memory* that captures the products of knowledge work in a community.  In a sense, the organizational memory is a "knowledge base" that lets people learn from and build on the work of their colleagues.  An implementation of this idea, the Intraspect Knowledge Server, has been successfully deployed and used by thousands of knowledge workers.  The key to its success has been that the memory is created as a natural byproduct of a community of people collaborating on-line, rather than as a centrally controlled activity of collecting and representing knowledge for consumption by a targeted audience.

For this seminar, we will discuss how this approach to "building a knowledge base" forces a new look at traditional questions of knowledge and representation: What are the requirements on knowledge representation when the creators and consumers are people?  What assumptions can be made about shared ontology?  What inferences and other cognitive functions can be supported by computation in an organizational memory?  What happens to the problems of "knowledge engineering" and "the knowledge acquisition bottleneck"?

About the Speaker:

Tom Gruber is cofounder and CTO of Intraspect, a leading vendor of collaborative knowledge management software.  From 1990-1994, Tom was a research associate at Stanford's Knowledge Systems, AI Laboratory.  At
Stanford, he worked on standards for sharing formally represented knowledge (ontologies) and created technologies for knowledge sharing among computer programs (Ontolingua).  When the Web arrived, he went off to work on knowledge sharing by people: first, by developing group collaboration technologies at EIT (Hypermail), and then to address the enterprise market at Intraspect.

Date: October 13, 1999

Speaker: Keith Edwards

Title: An Overview of the Jini Distributed Computing Infrastructure

Abstract:

The past few years have seen a dramatic increase in the number of devices and software systems that are "network-aware"--meaning that they provide some sort of way to connect to each other, and possibly to a range of network-hosted services.  As these systems become more complex, there is a danger that we will all, in a sense, become systems administrators:  tasked with maintaining and configuring a fragile distributed system, which may not be adaptable to changes in membership or the failure of individual components.

Sun's Jini platform is an attempt at addressing such issues.  In the Jini model, "services"--whether hardware or software--can be installed easily, and can interconnect with each other dynamically.  Collections of Jini services can form spontaneous "communities" of services that adapt to the addition--or failure--of members.

In this talk, I'll present an overview of the Jini technology, including its goals, its programming models, and some of its potential drawbacks.

Date: October 20, 1999

Speaker: Ramanathan Guha

Title: From a Web of Documents to a Web of Knowledge

Abstract:

The web today is a large document repository.  Very little of the information on the web is machine understandable. It is gradually moving towards becoming a repository of data. The real challenge is to take if from being a repository of data to a Web of Knowledge. In this talk, I will describe some of the infrastructure that is being build, such as RDF and the Open Directory, that might enable us to reach this goal.
 

Date: October 27, 1999

Speaker: Professor Mike Genesereth

Title: Semantic Integration of Heterogeneous Databases: The Relational Logic Approach

Abstract:

In recent years there has been a marked increase in the amount of "structured" data available on the world's computer networks, and all indicators suggest that this trend will continue in the years to come. Unfortunately, accessing this information in an integrated way is complicated by "semantic heterogeneity" among the data sources, i.e. differences in their schemas and vocabulary.  In this presentation, we will look at this problem in detail.  We will see why relational algebra is inadequate for the task and why relational logic, being more expressive, solves the problem.  We will look at one particular data integration system, called Infomaster, and a variety of its applications.  Finally, we will discuss some of the practical problems in developing large scale "datawebs", and we will examine the prospects for building a fully integrated "World Information Network", essentially a World Wide Web for databases.

Date: November 3, 1999

Speaker: Canceled

Title:

Abstract:
 
 

Date: November 10, 1999

Speaker: Daniel G. Bobrow

Title: "Eureka, Using the Web as a Community Knowledge Medium"

Abstract:

Xerox has 25,000 customer service engineers who repair equipment in the field.  Six years ago we built a model-based diagnosis system to support them in this work.  It failed for interesting reasons.  As a result, we engaged with them to build a system that supported their knowledge-sharing practice.  This community knowledge system uses a
semi-structured knowledge-base of cases, distributed community authoring and warranting, and web distribution of updates to technicians who usually cannot be connected while they work.  This system has been very successful.  Our current focus is to design support for the continuing management of the quality of this knowledge base, and other "closed" collections of documents.

Date: November 17, 1999, 9:00 AM

Speaker: Dr. Doug Lenat

Title: E-CYC:  Common Sense Moves to the Web

Abstract:

Date: November 17, 1999

Speaker: Dr. Benjamin Grosof

Title: Business Rules for E-Commerce on the Web: Courteous Logic Programs in XML, for Contracts
           and Authorization

Abstract:

In Part I of the talk, we discuss the overall usefulness of logic programs in multi-agent e-commerce, and the challenges that arise there.  We survey where and why rules are useful in specific multi-agent e-commerce areas, informed by our industry experience in industrial (as opposed to academic) research.  We observe that logic programs, as a core declarative knowledge representation, has wide current presence (e.g., in SQL databases and many non-Prolog rule systems) and major representational advantages compared to other rule approaches (as well as compared to non-rule programming approaches). Communicating logic programs between multiple agents, however, raises several challenges not well met previously.  First, a vehicle is needed for translation between heterogeneous rule systems.  This includes syntax and parsing, as well as deep declarative semantics. Second, facilities for conflict handling are needed when merging and updating rule sets, e.g., when an agent imports/assimilates rules from other agents or when multiple business-domain experts specify and maintain rule sets.  A kind of conflict important in many applications involves mutual exclusions of a more general kind than simply between p and not-p: e.g., that only one value for a price discount be concluded, among k possible values.

In Part II of the talk, we discuss our work that responds to these opportunities and challenges.  We have developed a research prototype, called CommonRules, for conflict handling and inter-agent communication in logic programs, that complements a variety of existing rule execution engines.  (An alpha version of CommonRules was released 7/30/99 free on the Web.) We are piloting applications of CommonRules in multi-agent e-commerce, including: - multi-agent negotiations that include communicating, and sometimes revising, contractual agreements that describe bids for goods (and services). A bid contains a contractual agreement that specifies not just pricing and product attributes of a good, but also the more complex terms & conditions of the proposed deal. - security authorization policies that include delegations between multiple agents (e.g., with cryptographically-authenticated certificates).

These applications make use of CommonRules' features for: - agent communications, including translations of rules/facts via a common "knowledge sharing" interlingua in XML, based on logic programs: Business Rules Markup Language. - prioritized handling of conflicts, based on Courteous logic programs, a form of prioritized logic programs that can be tractably compiled (e.g., via a preprocessor) to (semantically equivalent) ordinary logic progams.  This includes the capability to specify and consistently enforce generalized mutual exclusions of the kind discussed
in Part I of the talk.

This work is being piloted largely in the $29 Million 3-year EECOMS project on inter-enterprise integration in manufacturing supply chains, by an industry consortium (including IBM, Baan, Boeing, TRW, and smaller companies) being matching-funded by the US Government. Our collaborators also include five universities and AT&T Research.

Date: November 24, 1999

Speaker: Cancelled

Title:

Abstract:
 
 

Date: December 1, 1999

Speakers:   Lawrence M. Fagan, M.D., Ph.D.
                    Stanford Medical Informatics

                    Dan Berrios, M.D., M.P.H.
                    VA Palo Alto Health Care System
                    Stanford Medical Informatics

Title: Research Directions in Electronic Medical Publishing

Abstract:

This talk covers recent research in the Medical Information Interface Group at Stanford Medical Informatics.  Two projects will be highlighted, one designed to help index medical textbooks, the other aimed at helping patients to find appropriate medical information from citation databases. The origin of these efforts go back to early Knowledge Systems, AI Laboratory development of the MYCIN project!

Medical publishers have traditionally developed electronic versions of their paper-based textbooks primarily by building CDs with boolean search capabilities. With considerable information packed into each page, search techniques that look for the co-occuring terms in a local region of a textbook get
poor results.  We will present an alternative approach to indexing and retrieving information from medical textbooks including approaches for semi-automated markup of the texts.  The semi-automated indexing work continues as the Ph.D. thesis research of Dan Berrios.

While publishers are bringing medical information online for providers, patients are turning towards the World Wide Web for information about diseases. Most of the general search engines return too many undifferentiated page references to be helpful.   Wanda Pratt developed an approach to dynamic categorization of citations to address this problem during her recent Ph.D. studies in Medical Information Sciences.