International Journal on Artificial Intelligence Tools

Graph Coloring Boid Swarm (GCBS) is the successful application of Reynolds' Boid Swarm to solve the graph coloring problem, following appropriate mapping of the problem into boids' behaviors and interpretation of the global swarm states as problem solutions. A population P of boids moves in a closed torus-shaped space S. Each individual boid perceives a disk of radius R of its surrounding space, being able to exchange information with other boids lying inside this area of perception. Two mutually perceiving boids exchanging information are connected. In this paper we show that the ratio of the radius of perception to the space size can be critical for the convergence of the Boid Swarm to the optimal configurations. First, we derive the Percolation threshold for Elementary Boid Swarms (EBS), whose dynamics are the aggregation of the boids into spatial clusters. When the radius of perception is above this percolation threshold for Elementary Boid Swarms (EBS), whose dynamics are the aggregation of the boids into spatial clusters. When the radius of perception is above this percolation threshold there is complete connectivity and convergence to a single cluster. Second, we show empirically the existence of such a Percolation effect on the Graph Coloring Boid Swarm (GCBS). We find that complete connectivity has an adverse effect on the performance of the GCBS. Finally we show a comparison of GCBS with some other algorithms over a family of generated graphs.

Boolean Satisfiability (SAT) solvers have been successfully applied to a wide range of practical applications, including hardware model checking, software model finding, equivalence checking, and planning, among many others. SAT solvers are also the building block of more sophisticated decision procedures, including Satisfiability Modulo Theory (SMT) solvers. The large number of applications of SAT yields ever more challenging problem instances, and motivate the development of more efficient algorithms. Recent work studied hybrid approaches for SAT, which involves integrating incomplete and complete SAT solvers. This paper proposes a number of improvements to hybrid SAT solvers. Experimental results demonstrate that the proposed optimizations are effective. The resulting algorithms in general perform better and, more importantly, are significantly more robust.

Mining frequent tree patterns from tree databases has practical importance in domains like Web mining, Bioinformatics, and so on. Although there have been algorithms on efficient tree mining, these algorithms are often lack of the interpretability in that they often produce a huge number of patterns, most of which are meaningless to users. This paper aims at both demands, one with respect to computational cost, which is efficient generation of tree patterns, and another one with respect to the interpretability. This task requires an efficient method to incorporate the users' needs into mining process.

We propose a new top-down method for mining unordered closed tree patterns from a database of trees such that every mined pattern must contain a common piece of information in the form of a tree specified by the user. This type of mining is called mining with subtree constraint which would be useful, for example, inWeb mining and Bioinformatics, where users want to extract common patterns around some given information from original data. The proposed algorithm is tested and compared with a state-of-the-art tree mining algorithm on real and artificial datasets with very good results.

In this paper a new Genetic Algorithm is proposed, called GAppi, which performs clustering in protein-protein interaction networks to identify protein complexes. The algorithm has been tested exhaustively with experimental datasets coming from online protein interaction databases and individual experiments and it has been compared with five other effective techniques in order to demonstrate its efficiency and superior performance. Results showed that GAppi produces feasible and very efficient solutions compared to other techniques. Except from that, due to its adaptive behavior, each time it is used it can satisfy different constraints, thus meeting the different needs of each user. Furthermore, a user friendly interface has been implemented that hosts the proposed algorithmic strategy.

Artificial immune system is inspired by the natural immune system for solving computational problems. The immunological principles that are primarily used in artificial immune systems are the clonal selection principle, the immune network theory, and the negative selection mechanism. These principles have been applied in anomaly detection, pattern recognition, computer and network security, dynamic environments and learning, robotics, data analysis, optimization, scheduling, and timetabling. This paper describes how these three immunological principles were adapted by previous researchers in their artificial immune system models and algorithms. Finally, the applications of various artificial immune systems to various domains are summarized as a time-line.

Information extraction out of web pages, commonly known as screen scraping, is usually performed through wrapper induction, a technique that is based on the internal structure of HTML documents. As such, the main limitation of these kinds of techniques is that a generated wrapper is only useful for the web page it was designed for. To overcome this, in this paper it is proposed a system that generates first-order logic rules that can be used to extract data from web pages. These rules are based on visual features such as font size, elements positioning or types of contents. Thus, they do not depend on a document's internal structure, and are able to work on different sites. The system has been validated on a set of different web pages, showing very high precision and good recall, which validates the robustness and the generalization capabilities of the approach.

Collaborative Filtering (CF) is a popular technique employed by Recommender Systems, a term used to describe intelligent methods that generate personalized recommendations. Some of the most efficient approaches to CF are based on latent factor models and nearest neighbor methods, and have received considerable attention in recent literature. Latent factor models can tackle some fundamental challenges of CF, such as data sparsity and scalability. In this work, we present an optimal scaling framework to address these problems using Categorical Principal Component Analysis (CatPCA) for the low-rank approximation of the user-item ratings matrix, followed by a neighborhood formation step. CatPCA is a versatile technique that utilizes an optimal scaling process where original data are transformed so that their overall variance is maximized. We considered both smooth and non-smooth transformations for the observed variables (items), such as numeric, (spline) ordinal, (spline) nominal and multiple nominal. The method was extended to handle missing data and incorporate differential weighting for items. Experiments were executed on three data sets of different sparsity and size, MovieLens 100k, 1M and Jester, aiming to evaluate the aforementioned options in terms of accuracy. A combined approach with a multiple nominal transformation and a "passive" missing data strategy clearly outperformed the other tested options for all three data sets. The results are comparable with those reported for single methods in the CF literature.

Power plants are high complexity systems running risks of low frequency but high consequence. The field of machine learning appears to offer the necessary tools for developing automated instrument surveillance systems supporting decision-making in critical systems such as power stations. A novel prediction method is presented with the aim to enhance system safety and performance by making an ahead-of-time prediction of the status of fundamental system components and subsequent detection of abnormalities. The utilization of a linear assembly of support vector regressors employing unique kernels is proposed in a hybrid computational scheme that encompasses the formulation of a multi-objective optimization problem addressed with an evolutionary algorithm that employs Pareto theory to identify an optimal solution. The approach is tested on the ahead of time prediction of the crack length in power plant turbine blades utilizing historical data. The results obtained highlight the efficiency of the proposed methodology since better performance over the standalone support vector regressors is observed.

Visiting a foreign country is now much easier than it was in the past. This has led to a consequent increase in the need for translation services during these visits. To satisfy this need, a reliable translation assistance system based on sentence retrieval techniques is proposed. When a user inputs a sentence in his/her native language, the proposed system retrieves sentences similar to the input sentence from a pre-constructed bilingual corpus and returns pairs of sentences in the native and foreign languages. To reduce the lexical disagreement problems that inevitably occur in this sentence retrieval application, the proposed system uses multi-level linguistic information (i.e., keywords, sentence types, and concepts) with different weights as indexing terms. In addition, the proposed system uses clustering information from sentences with similar meanings to smooth the retrieval target sentences. In an experiment, the proposed system outperformed traditional IR systems. Based on various experiments, it was found that multi-level information was effective at alleviating critical lexical disagreement problems in sentence retrieval. It was also found that the proposed system was suitable for sentence retrieval applications such as translation assistance systems.

In this paper we present a novel method that forms a weighted combination of a range of Stacking based methods for regression problems, without adding any major computational overhead in comparison to stacking itself. The intention of the technique is to benefit from the variation in performance of individual Stacking methods as demonstrated with different data sets, in order to provide a more robust technique overall. We detail an empirical analysis of the technique referred to as weighted Meta–Combiner (wMetaComb) and compare its performance to its underlying techniques.