International Journal on Artificial Intelligence Tools

SAFIPS (Software Agents for IP Selection) environment is designed to improve quality and speed of finding IPs on Internet for SoC (System-on-Chip) projects. The software agents take responsibility for establishing communication with multiple suppliers of IPs and automatically retrieve information from their databases using metalanguage (XML), and data specifications compliant with ECIX (Electronic Component Information eXchange) specifications. The communication environment is based on Dictionaries for terminology and list of parameters, and on Registry for the list and web addresses of suppliers. The information obtained by software agents is analyzed by a system of fuzzy logic rules compiled in the process of a dialog between the customer and SA. The software agent asks the customer questions and provides templates for answers. The answers are automatically converted into the membership functions and fuzzy logic rules that are applied to evaluation of potential candidate IPs. As soon as replies to the query by SA start to come, data are analyzed by software agents and IPs can be excluded from further consideration if there is drastic mismatch between expected values and IP parameters and constraints. The remaining IPs are evaluated by the hierarchical system of rules and ranks are assigned to those which passed the preliminary tests. The top candidates are subjected to simulation together with the behavioral models provided by customers under supervision of software agent. The timing diagrams obtained from simulation are going to be analyzed to evaluate their equivalence. Only after that the system presents the candidate IPs for the final selection. The main goal to save valuable design time can be achieved by such system. The text is accompanied by an example.

In this paper, we present a new tool for solving weakly-coupled Markov Decision Processes using decomposition techniques. Using a predefined partition of the MDP, a directed graph is built to decompose the global MDP into small local MDPs which are independently solved. An approximate solution for the global MDP is obtained by combining local solutions. Our approach has been tested on a mobile robotics application. It allows near-optimal solutions to be obtained in significantly reduced time. We also present preliminary results concerning a parallel implantation of our tool.

Neural Networks are massively parallel processing systems, that require expensive and usually not available hardware, in order to be realized. Fortunately, the development of effective and accessible software, makes their simulation easy. Thus, various neural network's implementation tools exist in the market, which are oriented to the specific learning algorithm used. Furthermore, they can simulate only fixed size networks. In this work, we present some object-oriented techniques that have been used to defined some types of neuron and network objects, that can be used to realize, in a localized approach, some fast and powerful learning algorithms which combine results of the optimal filtering and the multi-model partitioning theory. Thus, one can build and implement intelligent learning algorithms that face both, the training as well as the on-line adjustment of the network size. Furthermore, the design methodology used, results to a system modeled as a collection of concurrent executable objects, making easy the parallel implementation. The whole design results in a general purpose tool box which is characterized by maintainability, reusability, and increased modularity. The provided features are shown by the presentation of some practical applications.

We propose an object oriented model for the simulation of large neural networks using the OMT technique. This modeling has been implemented on a client-server parallel simulator (called NeuSim-NNLIB), the server being a "beowulf" cluster, getting up to 18 MCPS with a cluster of 6 Pentium processors. We also present an estimation for simulator performance and optimal processors number.

The extension of logic programming with abduction (ALP) allows a form of hypothetical reasoning. The advantages of abduction lie in the ability to reason with incomplete information and the enhancement of the declarative representation of problems. On the other hand, concurrent logic programming is a framework which explores AND-parallelism and/or OR-parallelism in logic programs in order to efficiently execute them on multi-processor / distributed machines.

The aim of our work is to study a way to model abduction within the framework of concurrent logic programming, thus taking advantage of the latter's potential for parallel and/or distributed execution. In particular, we describe Abductive Pandora, a syntactic sugar on top of the concurrent logic programming language Pandora, which provides the user with an abductive behavior for a concurrent logic program. Abductive Pandora programs are then transformed into Pandora programs which support the concurrent abductive behavior through a simple programming technique while at the same time taking advantage of the underlying Pandora machine infrastructure.

Many new approaches to solve optimization problems have been proposed with the appearance of methods based on analogies to natural processes. For the user in practice, it is not only difficult to assess which method is the "best" one but also to judge how easy or complicated it is to implement and apply the methods. This is mainly due to the fact that most methods have interacting parameters which need to be tuned before a method can be used. The paper discusses ease of use and implementation as well as performance aspects based on computational tests. Thre iterative improvement methods (simulated annealing, tabu search, threshold accepting) and two evolutionary methods (genetic algorithms, PRSA) are compared with regard to an optimization problem (standard cell placement in chip design). Based on results from a large number of test runs, we discuss how easy or difficult it is to prepare the methods for a specific application. In summary, iterative improvement methods were found easier to implement and to use than evolutionary methods. They also outperformed the latter ones in terms of solution quality and computing time.

Magnetic flow meters (magmeters) are instruments for measuring the velocity of flow in many industrial applications. The signal that comes from a magmeter is noisy and conventional approaches are often not effective enough in dealing with actual field noise. Furthermore, diagnostic functions are attracting increasing attention, due to the possibility of implementing them in an inexpensive and reliable manner in magmeter hardware. Neural networks have proven capabilities for both learning and data handling in noisy circumstances. In this paper, a novel approach based on wavelet neural networks is presented to attack these two objectives. The stability, accuracy and response time of the new approach has been tested, and found to be superior to conventional approaches.

The vehicle routing problem with time windows (VRPTW) is an extension of the well-known vehicle routing problem (VRP). It involves a fleet of homogeneous vehicles, originating and terminating at a central depot, with limited capacity and maximum travel time to service a set of customers with known demands and service-time windows. The objective is to find a set of feasible routes that minimizes the total costs using some measures of solution quality. This paper focuses on the study of a hybrid of two search heuristics, Tabu Search (TS) and Genetic Algorithm (GA) on VRPTW. TS is a local search technique that has been successfully applied to many NP-complete problems. On the other hand, GA which is capable of searching multiple search areas in a search space is good in diversification. In this paper, we create a hybrid that combines the strengths of the two search heuristics. Experimental results indicate that such a hybrid outperforms the individual heuristics alone.