International Journal on Artificial Intelligence Tools

The adaptation of software technology to distributed environments is an important challenge today. In this work we combine parallel and distributed search. By this way we add the potential speed-up of a parallel exploration in the processing of distributed problems. This paper extends DIBT, a distributed search procedure operating in distributed constraint networks.¹¹ The extension is threefold. First, the ordered hierarchies used during backtracking are extended to remove partial orders. Second the procedure is updated to face delayed information problems upcoming in heterogeneous (We would like to thank here M. Yokoo for rising this potential problem.)Third, the search is extended to simultaneously explore independent parts of a distributed search tree. The first and third points were first presented in 1999.⁷ The third improvement introduces parallelism into distributed search, which brings to Interleaved Distributed Intelligent BackTracking (IDIBT). Our results show that 1) insoluble problems do not greatly degrade performance over DIBT and 2) super-linear speed-up can be achieved when the distribution of solutions is non-uniform.

Neural networks with a single hidden layer are known to be universal function approximators. However, due to the complexity of the network topology and the nonlinear transfer function used in computing the hidden unit activations, the predictions of a trained network are difficult to comprehend. On the other hand, predictions from a multiple linear regression equation are easy to understand but are not accurate when the underlying relationship between the input variables and the output variable is nonlinear. We have thus developed a method for multivariate function approximation which combines neural network learning, clustering and multiple regression. This method generates a set of multiple linear regression equations using neural networks, where the number of regression equations is determined by clustering the weighted input variables. The predictions for samples of the same cluster are computed by the same regression equation. Experimental results on a number of real-world data demonstrate that this new method generates relatively few regression equations from the training data samples. Yet, drawing from the universal function approximation capacity of neural networks, the predictive accuracy is high. The prediction errors are comparable to or lower than those achieved by existing function approximation methods.

In this paper we present a new method that uses data-flow coherence constraints in definite logic program generation. We outline three main advantages of these constraints supported by our results: i) drastically pruning the search space (around 90%), ii) reducing the set of positive examples and reducing or even removing the need for the set of negative examples, and iii) allowing the induction of predicates that are difficult or even impossible to generate by other methods. Besides these constraints, the approach takes into consideration the program termination condition for recursive predicates. The paper outlines some theoretical issues and implementation aspects of our system for automatic logic program induction.

The WEBSOM methodology for building very large text archives has a very slow method for extracting meaningful unit labels. This is due to the fact that the method computes for the relative frequencies of all the words of all the documents associated to each unit and then compares these to the relative frequencies of all the words of other units in the map. Since maps may have more than 100,000 units and the archieve may contain up to 7 million documents, the existing WEBSOM method is not practical. A fast alternative method, referred to as the liGHtSOM method, is based on the distribution of weights in the weight vectors of the trained map, plus a simple manipulation of the random projection matrix used for input data compression. Comparison made using a WEBSOM archieve of the Reuters text collection reveal that a high percentage of keywords extracted using this method match the keywords extracted for the same map units using the original WEBSOM method. A detailed time complexity analysis of the two methods is also provided.

We consider applications where agents have to cooperate without any communication taking place between them, apart from the fact that they can see part of the environment in which they act. We present a multi-agent system, defined in Golog, that needs to service tasks whose value degrades in time. Initial plans, reflecting prior knowledge about the environment, are expressed as Golog procedures, and are provided to the agents. Then the agents are trained using reinforcement learning, in order to ensure coordination both at the action level and at the plan level. This ensures better scalability and increased performance of the system.

Several real world applications involve solving combinatorial optimization problems. Commonly, existing heuristic approaches are designed to address specific difficulties of the underlying problem and are applicable only within its framework. We suspect, however, that search spaces of combinatorial problems are rich in intuitive statistical and numerical information, which could be exploited heuristically in a generic manner, towards achievement of optimized solutions. Our work presents such a heuristic methodology, which can be adequately configured for several types of optimization problems. Experimental results are discussed, concerning two widely used problem models, namely the Set Partitioning and the Kanpsack problems. It is shown that, by gathering statistical information upon previously found solutions to the problems, the heuristic is able to incrementally adapt its behaviour and reach high quality solutions, exceeding the ones obtained by commonly used greedy heuristics.

Nowadays, there is no doubt that machine learning techniques can be successfully applied to data mining tasks. Currently, the combination of several classifiers is one of the most active fields within inductive machine learning. Examples of such techniques are boosting, bagging and stacking. From these three techniques, stacking is perhaps the less used one. One of the main reasons for this relates to the difficulty to define and parameterize its components: selecting which combination of base classifiers to use, and which classifier to use as the meta-classifier. One could use for that purpose simple search methods (e.g. hill climbing), or more complex ones (e.g. genetic algorithms). But before search is attempted, it is important to know the properties of the search space itself. In this paper we study exhaustively the space of Stacking systems that can be built by using four base learning systems: C4.5, IB1, Naive Bayes, and PART. We have also used the Multiple Linear Response (MLR) as meta-classifier. The properties of this state-space obtained in this paper will be useful for designing new Stacking-based algorithms and tools.

A general approach to implement propagation and simplification of constraints consists of applying rules over these constraints. However, a difficulty that arises frequently when writing a constraint solver is to determine the constraint propagation algorithm. In previous work, different methods for automatic generation of rule-based solvers for constraints defined over finite domains have been proposed^1,2,3,4. In this paper, we present a method for generating rule-based solvers for constraint predicates defined by means of a constraint logic program, even when the constraint domain is infinite.