Catálogo de publicaciones - libros

The problem of ensuring agents work as an effective team in dynamic distributed environments still remains a challenging issue. In this paper we proposed a role-based team model. In our model, the role characterizes the responsibilities and provides logic patterns to achieve certain goals and cooperate with others. The agent is an autonomous execution unit and follows the logic patterns that the role provides. We also developed algorithms and mechanisms to evolve the plan of a role to the plan of an agent. Our role-based team model allows the split of roles (who define the plans) and agents (who execute the plans) in team plans, and dynamic role-agent assignment. It also achieves a certain level of plan reusability. We present two experiments which show plan reusability and its flexibility in supporting simultaneously plan invocation.

Recommender systems enable a user to decide which information is interesting and valuable in our world of information overload. Collaborative Filtering (CF), one of the most successful technologies in recommender systems suffers from improper use of personal information and the incredibility of recommendations. To deal with these issues, we have been focusing on the trust relationships between individuals, i.e. , especially for protecting the recommender system against profile injection attack. Based on trust propagation scheme, we proposed architecture which is added agent-based scheme obtaining attack resistance property as well as improving the efficiency of distributed computing. In , users’ personal agents find a unique migration path made up of latent neighborhoods and reduce search scope to a reasonable level for mobile agents by using the algorithm. The experimental evaluation on datasets shows that the proposed method brings significant advantages in terms of dealing with profile injection attack without any loss of prediction quality.

Agent competition and coordination are two classical and most important tasks in multiagent systems. In recent years, there was a number of learning algorithms proposed to resolve such type of problems. Among them, there is an important class of algorithms, called adaptive learning algorithms, that were shown to be able to converge in self-play to a solution in a wide variety of the repeated matrix games. Although certain algorithms of this class, such as Infinitesimal Gradient Ascent (IGA), Policy Hill-Climbing (PHC) and Adaptive Play -learning (APQ), have been catholically studied in the recent literature, a question of how these algorithms perform versus each other in general form stochastic games is remaining little-studied. In this work we are trying to answer this question. To do that, we analyse these algorithms in detail and give a comparative analysis of their behavior on a set of competition and coordination stochastic games. Also, we introduce a new multiagent learning algorithm, called ModIGA. This is an extension of the IGA algorithm, which is able to estimate the strategy of its opponents in the cases when they do not explicitly play mixed strategies (e.g., APQ) and which can be applied to the games with more than two actions.

Real-time vehicle location information enables to facilitate more efficient decision-making in dynamic automotive shipment yard environment. This paper proposes a multiagent-based decentralized decision-making model for the vehicle deployment planning in a shipment yard. A multiagent architecture is designed to facilitate decentralized algorithms and coordinate different agents dynamically. The results of computational experiments show that the proposed deployment model outperforms a current deployment practice with respect to the deployment performance measures.

Resource allocation is a widely studied class of problems in Operation Research and Artificial Intelligence. Specially, constrained stochastic resource allocation problems, where the assignment of a constrained resource do not automatically imply the realization of the task. This kind of problems are generally addressed with Markov Decision Processes (s). In this paper, we present efficient lower and upper bounds in the context of a constrained stochastic resource allocation problem for a heuristic search algorithm called Focused Real Time Dynamic Programming (). Experiments show that this algorithm is relevant for this kind of problems and that the proposed tight bounds reduce the number of backups to perform comparatively to previous existing bounds.

MAS failures are not only due to programming exceptions; they may originate from other sources such as the environment of the MAS which may influence the MAS’ behavior. Furthermore, MAS fault-tolerant techniques based on agent replication cannot always be applied to a MAS. For example, it is not always possible to replicate a costly robot in a robotic MAS application. In this paper, we propose a reorganization strategy, based on task and agent replication, to enable a MAS to detect and recover from its failures. Our strategy is different from those presented in the literature, which are based on agent replication, since it does not deal with programming faults but with failures originating from the MAS environment, and it is based on task and agent replication and not only on agent replication. Our strategy is scalable and is robust in detecting agents failure.

In this work, we build a decision tree to predict fungal protein localization based on physiochemical properties of proteins calculable from their primary sequences. Although there is clear evidence of presence of the same protein in more than one sub-cellular compartment, almost all existing automated systems restrict their predictions to single-site localization. We address this issue by predicting as many localization sites as possible. When localizing among 17 sub-cellular compartments, in 64% of the cases our system successfully predicts at least one of the experimentally reported localizations. Moreover, all reported localizations are correctly predicted in 49% of the cases. We also report 76 fungal protein features expected to be implicated in localization, based on the constructed decision tree.

Based on the general framework of logical analysis of data, we develop a probe design method for selecting short oligo probes for genotyping applications in this paper. When extensively tested on genomic sequences downloaded from the Lost Alamos National Laboratory and the National Center of Biotechnology Information websites in various monospecific and polyspecific experimental settings, the proposed probe design method selected a small number of oligo probes of length 7 or 8 nucleotides that perfectly classified all unseen testing sequences. These results well illustrate the utility of the proposed method in genotyping applications.

Many researchers have used text classification method in solving the ontology mapping problem. Their mapping results heavily depend on the availability of quality exemplars used as training data. However, manual preparation of exemplars is costly. In this work, we propose to automatically extract text from web pages returned by a search engine. Search queries are formed according to the semantic information given in the ontology. We have implemented a prototype system that automates the entire process (from search query formation to conditional probability calculation) and conducted a series of experiments. We assessed the effectiveness of our approach by comparing the obtained conditional probabilities with human expectations. Our main contribution is that we explored the possibilities of utilizing web information for text classification based ontology mapping and made several valuable discoveries on its usefulness for future research.

For high-dimensional classification tasks, such as face recognition, the number of samples is smaller than the dimensionality of the samples. In such cases, a problem encountered in Linear Discriminant Analysis-based (LDA) methods for dimension reduction is what is known as the Small Sample Size (SSS) problem. A number of LDA-extension approaches that attempt to solve the SSS problem have been proposed in the literature. Recently, a different way of employing a dissimilarity representation method was proposed [18], where an object was represented based on the dissimilarity measures among representatives extracted from training samples instead of the feature vector itself. Apart from utilizing the dissimilarity representation, in this paper, a new way of employing a fusion technique in representing features as well as in designing classifiers is proposed in order to increase the classification accuracy. The proposed scheme is completely different from the conventional ones in terms of the computation of the transformation matrix as well as the selection of the number of dimensions. The present experimental results demonstrate that the proposed combining mechanism works well and achieves further improved efficiency compared with the LDA-extension approaches for well-known face databases involving AT&T and Yale databases. The results especially demonstrate that the highest accuracy rates are achieved when the combined representation is classified with the trained combiners.