International Journal on Artificial Intelligence Tools

Synapses are key elements in the organization of neuronal circuits. To analyse these structures, it is now possible to obtain serial images from large samples of nervous tissue in an automated manner through combined focused ion beam milling and scanning electron microscopy. However, the identification, 3D reconstruction and quantification of synapses within these samples are labor intensive procedures that require continuous user intervention. We have developed a software tool to achieve the segmentation of synapses in a reconstructed 3D volume of the cerebral cortex, thereby greatly facilitating and accelerating these processes. The tool is interactive, allowing the user to supervise the segmentation process, to modify the appropriate parameters and to validate the results. Some experimental results obtained with our tool in the rat cerebral cortex are also presented.

Chemical reactions always involve several molecules of two types, reactants and products. Existing data mining techniques, eg. Quantitative Structure Activity Relationship (QSAR) methods, deal with individual molecules only. In this article, we propose to use a Condensed Graph of Reaction (CGR) to merge all molecules involved in a reaction into one molecular graph. This allows one to consider reactions as pseudo-molecules and to develop QSAR models based on fragment descriptors. Then ISIDA (In SIlico Design and Analysis) fragment descriptors built from CGRs are used to generate models for the rate constant of S_N2 reactions in water, using three usual attribute-value regression algorithms (linear regression, support vector machine, and regression trees). This approach is compared favorably to two state-of-the-art relational data mining techniques.

An agent-based Virtual Organization is a complex entity where dynamic collections of agents agree to share resources in order to accomplish a global goal or offer a complex service. An important problem for the performance of the Virtual Organization is the distribution of the agents across the computational resources. The final distribution should provide a good load balancing for the organization. In this article, a genetic algorithm is applied to calculate a proper distribution across hosts in an agent-based Virtual Organization. Additionally, an abstract multi-agent system architecture which provides infrastructure for Virtual Organization distribution is introduced. The developed genetic solution employs an elitist crossover operator where one of the children inherits the most promising genetic material from the parents with higher probability. In order to validate the genetic proposal, the designed genetic algorithm has been successfully compared to several heuristics in different scenarios.

When interacting with constraint-based recommender applications, users describe their preferences with the goal of identifying the products that fit their wishes and needs. In such a scenario, users are repeatedly adapting and changing their requirements. As a consequence, situations occur where none of the products completely fulfils the given set of requirements and users need a support in terms of an indicator of minimal sets of requirements that need to be changed in order to be able to find a recommendation. The identification of such minimal sets relies heavily on the existence of (minimal) conflict sets. In this paper we introduce BFX (Boosted FastXplain), a conflict detection algorithm which exploits the basic structural properties of constraint-based recommendation problems. BFX shows a significantly better performance compared to existing conflict detection algorithms. In order to demonstrate the performance of BFX, we report the results of a comparative performance evaluation.

Mobile robots have a large application potential in everyday life. To build those applications some common and basic behaviors should be initially consolidated, including a people following behavior. In this paper a system able to follow a person based on information provided by a laser scan and a mono and stereo camera is presented. In order to accomplish this goal, a real-time particle filter system able to merge the information provided by the sensors (laser and 2D and 3D images) and calculate the position of the target is proposed, using probabilistic leg patterns, image features and optical flow to this end. The experiments carried out show promising results, allowing a real-time particle filtering based on two different information sources.

For improving the interaction between students and teachers, it is fundamental for teachers to understand students' learning levels. An intelligent computer system should be able to automatically evaluate students' answers when the teacher asks some questions. We first built the assessment corpus from the course in the university. With the corpus, we applied the following procedures to extract the relevant information and then built the feature model: (1) remove the punctuation and decimal numbers because it plays the noise roles, (2) apply the part-of-speech tagging such that the syntactic information is extracted, (3) for grouping the information, take the stemming and normalization procedure to sentences, and (4) extract other features. In this study, we treated the assessment problem as the classifying problem, and tried two kinds of classification strategies: two and three classifying classes. For two classes, we got an average of 66.3% precision rate at first. When adding n-gram concept to the feature model, the system reached to the average of 71.9% precision rate which increased performance by 5.6%. The same tendency emerged for three-class experiment. The experiments with SVM show exhilarating results and some improving efforts will be further made in the future.

In a complex crime scene with many possible suspects and conflicting evidence, crime investigation requires scientific and logical steps to narrow down the suspects. Since human investigators have difficulty in fully handling all reasoning in this highly complex hypothesis space, we propose a decision support system to aid the investigation process. The system integrates a rule-based reasoner, a Bayesian network for criminal profiling, and an assumption-based truth maintenance system for evidential reasoning to determine the most plausible suspect. The reasoning process reasons about a profiling classification and an alibi credibility measure. It proved effective in a realistic simulation.

The importance of time series segmentation techniques is rapidly expanding, due to the growth in collection and storage technologies. Among them, one of the most used ones is Piecewise Linear Representation, probably due to its ease of use. This work tries to determine the difficulties faced by this technique when the analyzed time series shows noisy data and a large number of measurements and how to introduce the information about the present noise in the segmentation process. Both difficulties are met in the Air Traffic Control domain, which exhibits position measurements of aircraft's trajectories coming from sensor devices (basically surveillance radar and aircraft-derived data), being used as the motivating domain. Results from the three main traditional techniques are presented (sliding window, top down and bottom up approaches) and compared with a new introduced approach, the Hybrid Local Residue Analysis technique.