Please note, this page is out of date - click the link for Parka-DB below, for something more up to date (but even that page is getting world-weary)

Large-Scale Knowledge Representation:
The PARKA Project

Parallel Understanding Systems Group
Computer Science Department, University of Maryland at College Park


Screenshot of a PARKA-DB query (click for full scale picture)


PARKA-DB(TM) Winner of University of Maryland Invention of the Year!

Click here for more information on the integration of knowledge- and data-base technology!


The PARKA Project - An overview

The Parka system is a frame-based AI language (sometimes called a ``property/class'' system in todays literature) which was designed to be supported efficiently using high-performance computing techniques. The goal of the project is, and has always been, to develop a fairly traditional Artificial Intelligence language/tool that can scale to the extremely large size applications mandated by the needs of today's information technology revolution.

More specifically, Parka allows the user to define a frame-based knowledge base with class,subclass, and property links used to encode the ontology. Property values can themselves be frames, or alternatively can be string, numeric values, or specialized datastructures (used primarily in the implementation). The Parka language allows exceptions, in the form of multiple-inheritance, and provides extremely efficient (and efficiently parallelizable) algorithms for performing inheritance using a true inferential-distance-ordering calculation.

Parka has also been shown to effectively compute recognition, and also to handle extremely complex ``structure matching'' queries -- a class of conjunctive queries relating a set of variable and contraints and unifying these against the larger KB. While it is difficult to exactly compare KR languages, a very loose categorization would put Parka as more expressive than Classic due to the Parka's ability to handle exceptions, although slightly less expressive than Loom due to the lack of extensive numerical capabilities. (for papers on Classic, Loom and others, search the databasess and logic programming bibliography ). The manual for the Parka language and more details on past results can be found in our publications page .

One of the key features of Parka is that it has been shown to efficiently handle its inferencing on KBs containing millions of assertions. Early work on the system gained most of its efficiency through the use of massive parallelism , however in recent years we've made increasing use of database management techniques to remove the need for parallelism (although still allowing for efficient parallelization). The latest version of Parka uses DBMS technologies to support inferencing and data management. In particular, this system, called "Parka-DB" was developed to run on generic, single processor (or parallel) systems with significantly less primary memory requirements than the previous versions.

NOW AVAILABLE ON THE NET -- SEVERAL EXTREMELY LARGE KNOWLEDGE BASES you can download.

See the pages below for more information on Parka and on its applications,


PARKA Versions:

oPARKA-CM
oPARKA-DB
oHigh Performance PARKA


PARKA Applications:

o CaPER - A case-based planning system.
o Hybrid Knowledge Base / Database Support
o ForMAT with PARKA


o Parka Publications


Status/Availability:

If you are interested in using Parka please contact Prof. James Hendler at hendler@cs.umd.edu.


People:


Enhanced with SHOE