Integration of Tools for Intelligent Technologies

Supported by Ministry of Education, Slovak Republic, Grant No. 1/5032/98

Project Overview

The project concentrates on two ways of integration of tools developed in frame of different fields of Artificial Intelligence, these are: The project comprises also development of new or modified methods in different fields of AI with emphasis on possibility of their integration for complex tasks solving. The project will concentrate on Neural Networks, Evolutionary Algorithms, Artificial Life and Constraint Logic Programming.

Associated Faculty

Present state of the subject

In the field of Intelligent Technologies two big groups of means and methods have developed: These two different directions in AI have been for a long time presented as two isolated parts of AI and also inside these directions the research is done in a few considerably different fields. It is getting more and more obvious that for solution of complex tasks it is advantageous to make use of combined methods and procedures.

One of possible ways is to design an architecture and structure of communication among heterogeneous distributed agent-based tools of AI so as to secure the solution of highest possible quality in acceptable time period using up minimal computational resources

Design-to-time is an approach for solution of problems in fields, where the following conditions have been fulfilled: 1. there is not sufficient time to find a complete (or optimal) solution, 2. a time limit is defined for solution search, 3. there is more than one applicable methods distinguishing in quality of solution found and in the computation resources needed, 4. also non-optimal solution is acceptable, 5. it is possible to predict time and resources needed to find the solution, although the level of predictability for new tasks may be low

These characteristics are in many respects equal with real-time problems where there is not enough time to find optimal solution. Depending on time and available resources it is possible to choose a suitable means that will find a suboptimal solution under given conditions simultaneously minimizing the load on other resources.

Although different approaches for solution of this class of tasks have been applied (anytime algorithms, imprecise computation), there exists no general methodology yet. Approaches exploiting multi-agent technologies show up to be promising at this point.

Another way in exploiting seemingly completely heterogeneous methods of AI is to include different tools of intelligent technologies into one integral tool that will be applied for solution of real-life problem. Currently there is a few of such tools integrating different AI methods. One of interesting products of Siemens company is programmers version of integrating programming system ECANSE for operating system WINDOWS NT. This tool is suitable for integration of AI methods such as CLP method (Constraint Logic Programming), Neural Networks, Fuzzy Regulator, Genetic Algorithms etc. in order to solve given real-life problem.

In last few years constraints satisfying programming systems integrating very effective algorithms of propagation and constraints satisfaction are getting into the front and they present very interesting method for solving and prototyping of whole scale of real-life applications mainly because of their declarative characteristics and increasing efficiency. They represent modern programming systems based on constraint logic programming or object-oriented and distributed programming systems. Some of these systems in spite of their short history have become commercially very successful. But these systems are far too expensive regarding our current possibilities. There is also a group of systems available free of charge mostly for academic purposes and their performance is roughly comparable to the commercial systems. These systems present excellent development tool which is possible to exploit for pedagogical purposes as well as for purposes of development of different applications and testing of new methods. And just the AI tools designed using these development tools are suitable for inclusion into integrated or multi-agent environment.

Biologically motivated methods of AI represent one of key groups of tools we are intending to integrate in frame of this project. But these methods alone are often based on co-operation of huge amount of elements and due to emergency the integral tool obtains abilities exceeding the sum of abilities of the integrated elements. These phenomena are studied by Artificial Life (ALIFE).

Evolutionary algorithms present a class of searching algorithms inspired by natural processes. They apply population principle to searching for the solution. This principle enables concurrent searching through the space of potential solutions. Their properties predestine them for solving of tasks not requiring the guarantee of finding the optimal solution. Ability to solve also problems the complexity of which is beyond the frontiers of applicability of full algorithms along with potential to find alternative solutions and solutions based on more goals makes a promising means of solution of broad range of tasks from technical praxis out of this class of algorithms. Broader application of evolutionary algorithms is hindered by relatively complex process of design needed to create an efficient implementation. In the present a typical design process consists of creation of ad-hoc variant of the algorithm and its gradual empirical improvement selecting alternative parts of algorithm, setting more suitable values of control parameters, or implementing domain knowledge.

Particular contribution expected

We consider the main contribution of the project to be development of two types of environment allowing integration of heterogeneous methods of artificial intelligence: Development of integral system for application of different means of AI will also enable more effective utilization of means and methods of AI for solution of real-life problems. Also finding of solution of some partial tasks aiming at creation of AI tools suitable for inclusion into the above mentioned environments may be considered to be contribution of this project. These partial tasks include:

Proposal of the ways how to reach the project goals

Regarding development of multi-agent environment we propose following consecutive steps: Regarding development of integrated programming environment we plan to take following steps: In development of methods and tools based on Constraint Logic Programming we will go through following stages: In the field of design and application of evolutionary algorithms we will concentrate on gradual building up of suitable development environment for design of applications of evolutionary algorithms which would allow quick prototypical design with possibility of empirical evaluation of more variants of solution, of optimization of the performance of algorithm and consecutive generation of code for the target environment.

In study of artificial life and application of obtained knowledge we will proceed in following steps: