Constrainted Object Hierarchy - An Architecture for Intelligent Systems

Constrained Object Hierarchy

-An Architecture for Intelligent Systems

Hongxue Wang

Department of Computer Science The Australian National University ACT 0200, AUSTRALIA

Email: hongxue@cs.anu.edu.au

This thesis will report investigations into an archi­ tecture for intelligent systems and associated knowledge representation scheme. Such a system is an object con­ taining subobjects like itself, some actions which can be taken and a set of constraints which these must satisfy. The language for programming these systems is quite general, high level and declarative. Our implementation takes advantage of the naturally distributed architecture for concurrent computation.

Any computing system is a kind of abstraction of some part of the real world. Computer scientists have tradi­ tionally worked on how to make abstractions from the world, how to encode an abstraction into a computer system and how to manipulate the abstraction inside the system for desired purposes. As a result, special fea­ tures and micro-structures of many parts of the world are well understood, and many formalisms such as program­ ming languages and knowledge representation schemes have been invented, while many manipulation strategies such as those of calculation and inference have been de­ veloped. However, the general features and the macro­ structures of the real world have been less investigated and rarely used for abstraction- that is, for the develop­ ment of computing systems. This is the problem to be addressed in the thesis.

We begin from a generic framework, which we de­ velop by reflecting on generic aspects of the world and some theoretical and technical results in computer sci­ ence. The following contributions should be made by the thesis:

1. It will address difficulties arising in developing such a representation scheme and problem solving system;
2. The generic formalism will be widely applicable for ab­ straction and for computer-based intelligent system de­ velopment;
3. As normal, it will provide a mechanism whereby large scale computing systems can easily he developed from small ones;
4. Dishihuted and parallel computation can naturally he applied in such a computing system;
5. We may gain better understanding of the structure of real world problems on which our abstraction is based.

In our account, any part of the real world is treated as a constrained object hierarchy(COH), which is made from three kinds of components: relatively static objects, functional actions and relations between object and ob­ ject, action and action or object and action. Three ma­ jor issues arise. The particle problem is that of ontology: what ground objects, primitive actions and fundamental relations should be used? The construction problem is how to combine primitives to introduce new objects, ac­ tions and relations for abstracting a specific part of the world. Finally, the manipulation problem is how to de­ fine the computations needed within a constrained object hierarchy and to determine how these should be carried out.

Investigation of the first two problems has led to the integration of objects from OOP and constraints from CSP, with a mechanism for pushing constraints into ob­ ject hierarchies. It becomes evident that many differ­ ent computing systems can be easily formalised as con­ strained object hierarchies of this kind.

Resolving the third issue has produced a classification of problem solving tasks, a classification of the elemen­ tary computations which can be carried out within a constrained object hierarchy. Some strategies for con­ trolling various computations and their interactions have also been proposed.

As part of the research, an implementation of COH theory has been made for a parallel computer. It is being used to evaluate the correctness of the theory and the efficiency of some proposed computation strategies, and to develop algorithms such as for constraint resolution within object hierarchies.

This research cuts across many boundaries within the field of AI, having connections with knowledge represen­ tation, constraint-based reasoning, reactive systems and distributed AI. We aim at an enhanced declarative for­ malism for knowpedge and problem representation, at a powerful engine for solving a wide range of problems and at a generic architecture for developing intelligent systems.