Integrating Distributed Knowledge
using Vivid Agents

Michael Schroeder
City University, London, 
msch@cs.city.ac.uk, 
Tel.: ++44 171 477 8918

ABSTRACT: 

We use vivid agents to integrate distributed information.  We show how
the agents' reactive and pro-active behaviour can be specified to
answer queries with knowledge distributed over different knowledge
bases. Currently, we are working on an application in Bioinformatics
involving a variety of databases ranging from flat files to
object-oriented and including existing legacy systems operating on
these databases.


OVERVIEW:

Exploring the enormous wealth of information in the WWW is a challenge
for DAI, in general, and agent technology, in particular. Finding
desired information in the WWW is still an open problem. Search
engines or web directories depend heavily on the keywords used and
have the user combine information manually. A possible solution to
overcome this shortcoming uses information agents that know about
available information from content providers.

Wiederhold and Genesereth point out several services that are
desirable for information agents. Many of them rising
problems not yet tackled nor solved.

1.  Determination of likely resources using information extracted earlier
2.  Invocation of wrappers deal with legacy sources
3.  Selection of relevant source material
4.  Optimization of access strategies to provide small response times or low cost
5.  Imposition of security filters to guard private data
6.  Resolution of domain terminology and ontology differences
7.  Resolution of scope mismatches
8.  Interpolation or extrapolation to match differences in temporal data
9.  Reduction of historical data to limited snapshots
10.  Abstraction to bring material to matching levels of granularity for integration 
11.  Integration of material from diverse source domains based on join keys
12.  Omission of replicated information
13.  Assessment of quality of material from diverse sources
14.  Pruning of data ranked low in quality or relevance
15.  Omission of information already known according to the customer model
16.  Statistical summarization into higher level objects, as defined in the customer model 
17.  Relaxation of search terms to satisfy query expectations
18.  Reporting exceptions from expected values or trends
19.  Triggering of actions due to exceptions from expected values or trends
20.  Transformation of material to make presentation effective for the customer
21.  adaption to the bandwidth  and media capabilities of the customer
22.  Sending the information and meta-information to the customer application


We are working on vivid agents and they may contribute to deal some of
the problems mentioned above.  A vivid agent is a software-controlled
system whose state is represented by a knowledge base, and whose
behaviour is represented by means of action and reaction rules.
Following Shoham, the state of an agent is described in terms of
mental qualities, such as beliefs and intentions.  The basic
functionality of a vivid agent comprises a knowledge system (including
an update and an inference operation), and the capability to represent
and perform actions in order to be able to generate and execute
plans. Since a vivid agent is `situated' in an environment with which
it has to be able to communicate, it also needs the ability to react
in response to perception events, and in response to communication
events created by the communication acts of other agents.

Vivid agents may contribute to some of the open issues for exploration
of the WWW. Mediators may record user queries, extract interests and
maintain necessary data (1). The model of the environment allows the
mediator to request the appropriate sites (3). The deliberative
planner of the mediator can be used to generate optimal queries to
reduce response times and costs (4). The model of the reagents allows
to resolve different terminology of the particular reagents (6). The
knowledge is obtained from distributed sites (11). The knowledge base
of the mediator contains knowledge about how to integrate diverse data
to create homogeneous views so that data is presented effectively to
the user (20).

Currently, we are applying these ideas to the integration of multiple
databases (including flat files, relational and object-oriented) in
Bio-Informatics. Together with the European Bio-Informatics Institute
in Cambridge we are developing an agent systems that plans to annotate
existing entries in the databses, thus integrating the partial
knowledge.