Progress in Artificial Intelligence

A genetic algorithm was used to search for optimum calibration parameter values for the vehicle performance models used by two well-known microscopic traffic simulation models, CORSIM and Integration. The mean absolute error ratio between simulated and empirical performance curves was used as the objective function. Empirical data was obtained using differential GPS transponders installed on trucks travelling on divided highways in Brazil. Optimal parameter values were found for the “average” truck for each truck class and for each vehicle in the sample. The results clearly show the feasibility of the proposed approach. The simulation models calibrated to represent Brazilian trucks individually provided average errors of 2.2%. Average errors around 5.0% were found when using the average truck class parameters.

A framework based on the concept of service-oriented architectures (SOA) to support the assessment of vehicular

ad-hoc

networks (VANET) is herein presented. Concepts related to SOA, as well as technologies that allow real-time data acquisition and dissemination within urban environments, and simulation tools to aid the simulation of VANET were preliminarily studied. A two-layered architecture was specified on the basis of the requirements for our simulation framework resulting in the specification of a multi-agent system formed of vehicle entities that are able to communicate and interact with each other and with their surrounding environment as well. A prototypical application was implemented, which was used to demonstrate the feasibility of the approach presented through experimental results.

Sensor-based traffic management systems have to cope with a high volume of continuously generated events. Conventional software architectures do not explicitly target the efficient processing of continuous event streams. Recently, Event-Driven Architectures (EDA) have been proposed as a new paradigm for event-based applications. In this paper we propose a reference architecture for event-driven traffic management systems, which enables the analysis and processing of complex event streams in real-time. In particular we are going to outline the different stages of traffic event processing and present an approach based on event patterns to diagnose traffic problems. The usefulness of our approach has been proven in a real world traffic management scenario.

We investigate the aspects that influence the instability of spatial evolutionary games, namely the Prisoner’s Dilemma and the Snowdrift games. In this paper instability is defined as the proportion of strategy changes in the asymptotic period of the evolutionary process. The results show that with the Prisoner’s Dilemma, when the level of noise present in the decision process is very low, the instability decreases as the synchrony rate decreases. With the Snowdrift this pattern of behavior depends strongly on the interaction topology and arises only for random and scale-free networks. However, for large noise values, the instability in both games depends only on the proportion of cooperators present in the population: it increases as the proportion of cooperators approaches 0.5. We advance an explanation for this behavior.

The paper presents an agent based model of the

Aedes aegypti

mosquito and it is focused on simulations of population dynamics and population control strategies. The agents model the main aspects of mosquito’s ecology and behavior, while the environmental components are implemented as layer of dynamic elements obeying to physical laws. The main objective of this approach is to provide realistic simulations of insect biologic control strategies, namely population suppression by releasing large amounts of sterile males, such as Sterile Insect Technique (SIT) or Release of Insects carrying a Dominant Lethal gene (RIDL). Model verification is done through simulations analysis of parameters variation and qualitative assessment with existing models and simulations. The use of LAIS simulator proved to be a valuable tool allowing efficient agent based modeling (ABM) and simulations deployment and analysis.

Predicting the toxicity of new potential drugs is a fundamental step in the drug design process. Recent contributions have shown that, even though Genetic Programming is a promising method for this task, the problem of predicting the toxicity of molecular compounds is complex and difficult to solve. In particular, when executed for predicting drug toxicity, Genetic Programming undergoes the well-known phenomenon of bloat, i.e. the growth in code size during the evolutionary process without a corresponding improvement in fitness. We hypothesize that this might cause overfitting and thus prevent the method from discovering simpler and potentially more general solutions. For this reason, in this paper we investigate two recently defined variants of the operator equalization bloat control method for Genetic Programming. We show that these two methods are bloat free also when executed on this complex problem. Nevertheless, overfitting still remains an issue. Thus, contradicting the generalized idea that bloat and overfitting are strongly related, we argue that the two phenomena are independent from each other and that eliminating bloat does not necessarily eliminate overfitting.

Rapid advances in science and in laboratorial and computing methods are generating vast amounts of data and scientific literature. In order to keep up-to-date with the expanding knowledge in their field of study, researchers are facing an increasing need for tools that help manage this information. In the genomics field, various databases have been created to save information in a formalized and easily accessible form. However, human curators are not capable of updating these databases at the same rate new studies are published. Advanced and robust text mining tools that automatically extract newly published information from scientific articles are required. This paper presents a methodology, based on syntactic parsing, for identification of gene events from the scientific literature. Evaluation of the proposed approach, based on the BioNLP shared task on event extraction, produced an average F-score of 47.1, for six event types.

RNA secondary structure prediction depends on context. When only a few (sometimes putative) RNA homologs are available, one of the most famous approach is based on a set of recursions proposed by Sankoff in 1985. Although this modus operandi insures an algorithmically optimal result, the main drawback lies in its prohibitive time and space complexities. A series of heuristics were developed to face that difficulty and turn the recursions usable. In front of the inescapable intricacy of the question when handling the full thermodynamic model, we come back in the present paper to a biologically simplified model that helps focusing on the algorithmic issues we want to overcome. We expose our ongoing developments by using the constraints framework which we believe is a powerful paradigm for heuristic design. We give evidence that the main heuristics proposed by others (structural and alignment banding, multi-loop restriction) can be refined in order to produce a substantial gain both in time computation and space requirements. A beta implementation of our approach, that we named ARNICA, exemplify that gain on a sample set that remains unaffordable to other methods. The sources and sample tests of ARNICA are available at

http://centria.di.fct.unl.pt/~op/arnica.tar.gz

Logic Programming Update Languages were proposed as an extension of logic programming that allows modeling the dynamics of knowledge bases where both extensional (facts) and intentional knowledge (rules) may change over time due to updates. Despite their generality, these languages do not provide a means to directly access past states of the evolving knowledge. They are limited to so-called Markovian change, i.e. changes entirely determined by the current state.

We remedy this limitation by extending the Logic Programming Update Language EVOLP with LTL-like temporal operators that allow referring to the history of the evolving knowledge base, and show how this can be implemented in a Logic Programming framework.

The execution model on which most tabling engines are based allocates a choice point whenever a new tabled subgoal is called. This happens even when the call is deterministic. However, some of the information from the choice point is never used when evaluating deterministic tabled calls with batched scheduling. Moreover, when a deterministic answer is found for a deterministic tabled call, the call can be completed early and the corresponding choice point can be removed. Thus, if applying batched scheduling to a long deterministic computation, the system may end up consuming memory and evaluating calls unnecessarily. In this paper, we propose a solution that tries to reduce this memory and execution overhead to a minimum. Our experimental results show that, for deterministic tabled calls and tabled answers with batched scheduling, it is possible not only to reduce the memory usage overhead, but also the running time of the execution.

Prolog is the most well-known and widely used logic programming language. A large number of Prolog applications maintains information by

asserting

and

retracting

clauses from the database. Such dynamic predicates raise a number of issues for Prolog implementations, such as what are the semantics of a procedure where clauses can be retracted and asserted while the procedure is being executed. One advantage of Logical Update semantics is that it allows indexing. In this paper, we discuss how one can implement just-in-time indexing with Logical Update semantics. Our algorithm is based on two ideas: stable structure and fragmented index trees. By

stable structure

one means that we define a structure for the indexing tree that not change, even as we assert and as we retract clauses. Second, by

fragmented index tree

we mean that the indexing tree will be built in such a way that the updates will be local to each fragment. The algorithm was implemented and results indicate significant speedups and reduction of memory usage in test applications.

In this paper, we describe a novel approach to tackle intention recognition, by combining dynamically configurable and situation-sensitive Causal Bayes Networks plus plan generation techniques. Given some situation, such networks enable recognizing agent to come up with the most likely intentions of the intending agent, i.e. solve one main issue of intention recognition; and, in case of having to make a quick decision, focus on the most important ones. Furthermore, the combination with plan generation provides a significant method to guide the recognition process with respect to hidden actions and unobservable effects, in order to confirm or disconfirm likely intentions. The absence of this articulation is a main drawback of the approaches using Bayes Networks solely, due to the combinatorial problem they encounter.

Inductive Logic Programming (ILP) [1] systems are general purpose learners that have had significant success on solving a number of relational problems, particularly from the biological domain [2,3,4,5]. However, the standard compression guided top-down search algorithm implemented in state of the art ILP systems like Progol [6] and Aleph [7] is not ideal for the Head Output Connected (HOC) class of learning problems. HOC is the broad class of predicates that have at least one output variable in the target concept. There are many relevant learning problems of this class such as arbitrary arithmetic functions and list manipulation predicates which are useful in areas such as automated software verification[8]. In this paper we present a special purpose ILP system to efficiently learn HOC predicates.

Starting with currently available data sets that capture the relationships between different pairs of personality-related constructs (for instance, descriptions of behaviors and personality traits), we describe a way of representing personality traits, human values, beliefs, and behaviors in one integrated network-centric model. We propose that such a unified representation can contribute to the making of predictions from one type of construct to another, for example, between personality characteristics and value preferences. Using a spreading activation mechanism, we show how, in principle, such a model can be used to make inferences about the plausibility of unobserved characteristics of an individual, given a very limited initial set of known characteristics.

Although several formal models of emotions for intelligent agents have recently been proposed, such models often do not formally specify how emotions influence the behavior of an agent. In psychological literature, emotions are often viewed as heuristics that give an individual the tendency to perform particular actions. In this paper, we take an existing formalization of how emotions come about in intelligent agents and extend this with a formalization of action tendencies. The resulting model specifies how the emotions of an agent determine a set of actions from which it can select one to perform. We show that the presented model of how emotions influence behavior is intuitive and discuss interesting properties of the model.

The exhibition of information does not always attend to the preferences and characteristics of the users, nor the context that involves the user. With the aim of overcoming this gap, we propose an emotional context-aware model for adapting information contents to users and groups. The proposed model is based on OCC and Big Five models to handle emotion and personality respectively. The idea is to adapt the representation of the information in order to maximize the positive emotional valences and minimize the negatives. To evaluate the proposed model it was developed a prototype for adapting RSS news to users and group of users.

This paper addresses the problem of propagating constraints involving arbitrary algebraic expressions. We formally describe previous approaches to this problem and propose a new model that does not decompose the expression thus avoiding introducing auxiliary data structures. We show how this compilation model fits naturally in a popular programming language supporting type parametricity, yielding significant speedups with respect to previous models.

In this paper we study colored nonogram solving using Integer Linear Programming. Our approach generalizes the one used by Robert A. Bosch which was developed for black and white nonograms only, thus providing a universal solution for solving nonograms using ILP. Additionally we apply a known algorithm to find all solutions to a puzzle. This algorithm uses a binary cut to exclude already known solutions. Finally we compare the performance of our approach in solving colored nonograms against other approaches, namely the iterative and the brute-force ones, pointing to a research direction of developing a hybrid method combining the iterative approach with ILP.

Different types of visual object categories can be found in real-world applications. Some categories are very heterogeneous in terms of local features (broad categories) while others are consistently characterized by some highly distinctive local features (narrow categories). The work described in this paper was motivated by the need to develop representations and categorization mechanisms that can be applied to domains involving different types of categories. A second concern of the paper is that these representations and mechanisms have potential for scaling up to large numbers of categories. The approach is based on combinining global shape descriptors with local features. A new shape representation is proposed. Two additional representations are used, one also capturing the object’s shape and another based on sets of highly distinctive local features. Basic classifiers following the nearest-neighbor rule were implemented for each representation. A meta-level classifier, based on a voting strategy, was also implemented. The relevance of each representation and classifier to both broad and narrow categories is evaluated on two datasets with a combined total of 114 categories.

This paper proposes a classification approach to identify the team’s formation (formation means the strategical layout of the players in the field) in the robotic soccer domain for the two dimensional (2D) simulation league. It is a tool for decision support that allows the coach to understand the strategy of the opponent. To reach that goal we employ Data Mining classification techniques. To understand the simulated robotic soccer domain we briefly describe the simulation system, some related work and the use of Data Mining techniques for the detection of formations. In order to perform a robotic soccer match with different formations we develop a way to configure the formations in a training base team (FC Portugal) and a data preparation process. The paper describes the base team and the test teams used and the respective configuration process. After the matches between test teams the data is subjected to a reduction process taking into account the players’ position in the field given the collective. In the modeling stage appropriate learning algorithms were selected. In the solution analysis, the error rate (% incorrectly classify instances) with the statistic test t-Student for paired samples were selected, as the evaluation measure. Experimental results show that it is possible to automatically identify the formations used by the base team (FC Portugal) in distinct matches against different opponents, using Data Mining techniques. The experimental results also show that the SMO (Sequential Minimal Optimization) learning algorithm has the best performance.

This paper describes the strategy and implementation details of a cooperative agent team for the CiberMouse@RTSS08, a robotics simulation competition. The paper introduces several concepts concerning cooperative robotics, giving an overview of its applications and challenges. It also presents the CiberMouse@RTSS08 competition rules and the associated simulation system. The proposed approach is based on a deliberative architecture, thus providing an autonomous intelligent behaviour. A probabilistic mapping procedure, based on the Bayes’ theorem, is used to maintain an internal world state. Quadtrees and A* are used for path planning and plan execution. A specific methodology was developed for beacon finding. The agents cooperate on exchanging world and beacon information. In order to overcome the simulator’s broadcasting distance limitation, a communication network between the agents was created. The paper analyses the impact of cooperation on this particular problem, by comparing the team’s performance in four different situations: communication with network and beacon exchanging, communication without network, communication without beacon exchanging and no communication at all.

For robots to interact with humans at the language level, it becomes fundamental that robots and humans share a common language. In this paper, a social language grounding paradigm is adopted to teach a robotic arm basic vocabulary about objects in its environment. A human user, acting as an instructor, teaches the names of the objects present in their shared field of view. The robotic agent grounds these words by associating them to visual category descriptions. A component-based object representation is presented. An instance based approach is used for category representation. An instance is described by its components and geometric relations between them. Each component is a color blob or an aggregation of neighboring color blobs. The categorization strategy is based on graph matching. The learning/grounding capacity of the robot is assessed over a series of semi-automated experiments and the results are reported.

This paper describes the approach developed by the CAMBADA robotic soccer team to address physical constraints regarding omni-directional motion control, with special focus on system delay. CAMBADA robots carry inherent delays which associated with discrete time control results in non-instant, non-continuous control degrading the performance over time. Besides a natural maximum velocity, CAMBADA robots have also a maximum acceleration limit implemented at software level to provide motion stability as well as current peaks avoidance on DC motors. Considering the previous constraints, such as the cycle time and the overall sensor-action delay, compensations can be made to improve the robot control. Since CAMBADA robots are among the slowest robots in the RoboCup Medium Size League, such compensations can help to improve both several behaviours as well as a better field coverage formation.

This paper presents an approach to endow a humanoid robot with the capability of learning new objects and recognizing them in an unstructured environment. New objects are learnt, whenever an unrecognized one is found within a certain (small) distance from the robot head. Recognized objects are mapped to an ego-centric frame of reference, which together with a simple short-term memory mechanism, makes this mapping persistent. This allows the robot to be aware of their presence even if temporarily out of the field of view, thus providing a primary spatial model of the environment (as far as known objects are concerned). SIFT features are used, not only for recognizing previously learnt objects, but also to allow the robot to estimate their distance (depth perception). The humanoid platform used for the experiments was the iCub humanoid robot. This capability functions together with iCub’s low-level attention system: recognized objects enact salience thus attracting the robot attention, by gazing at them, each one in turn. We claim that the presented approach is a contribution towards linking a bottom-up attention system with top-down cognitive information.

In robotic soccer, teams of autonomous robots play soccer according to rules similar to the official FIFA rules. The game is refereed by a human and his orders are communicated to the teams using an application called “Referee Box”. No human interference is allowed during the games except for removing malfunctioning robots and re-entering robots in the game. The base station, a software application as described in this paper, has a determinant role during the development of a robotic soccer team and also during a game. This application must control the agents interpreting and sending high level instructions, like

Start

or

Stop

, and monitor information of the robots, for example the position and velocity, allowing easily to attest the feasibility of the robots behavior. This paper discusses the importance of the control and monitoring of a robotic soccer team, presenting the main challenges and the approaches that were used by the CAMBADA team in the conception of the base station application. As far as we know, no previous work has been published about the study of these important problems and the discussion of an efficient architecture to a base station application. The results obtained by the team confirms the good performance of this software, both during the games and in the development of the team.

An architecture for the assignment of tasks in a multirobot system is presented. The tasks arrival times and order are not known, and no statistical distributions are assumed. Furthermore, the execution of tasks cannot be preempted and hard deadlines are imposed on both their assignment and conclusion times. The architecture design is inspired by economic markets and a performance measure, similar to market prices, is introduced. A theoretical analysis of the architecture operation is presented and the results are applied to the dimensioning of an active surveillance system for the UPC campus at Barcelona, Spain.

This paper presents the team coordination methodologies of CAMBADA, a robotic soccer team designed to participate in the RoboCup middle-size league (MSL). The coordination model extends and adapts previous work in the Soccer Simulation League to the MSL environment. The approach is based on flexible positionings and priority-based dynamic role/positioning assignment. In addition, coordinated procedures for passing and setplays have been implemented. With the described design, CAMBADA reached the 1st place in the RoboCup’2008 world championship, becoming the first Portuguese real robot team to win in RoboCup. Competition results and performance measures computed from logs and videos of real competition games are presented and discussed.

Robots need to ground their external vocabulary and internal symbols in observations of the world. In recent works, this problem has been approached through combinations of open-ended category learning and interaction with other agents acting as teachers. In this paper, a complementary path is explored, in which robots also resort to semantic searches in digital collections of text and images, or more generally in the Internet, to ground vocabulary about objects. Drawing on a distinction between broad and narrow (or general and specific) categories, different methods are applied, namely global shape contexts to represent broad categories, and SIFT local features to represent narrow categories. An unsupervised image clustering and ranking method is proposed that, starting from a set of images automatically fetched on the web for a given category name, selects a subset of images suitable for building a model of the category. In the case of broad categories, image segmentation and object extraction enhance the chances of finding suitable training objects. We demonstrate that the proposed approach indeed improves the quality of the training object collections.

When building a representation of the environment for a robot in a multi-agent application, as is the case of robotic soccer, sensor and information fusion of several elements of the environment are an important task. To build an increasingly better world model, one of the aspects that one should consider is the treatment of obstacles. This paper gives an insight of the general steps necessary for a good obstacle representation in the robot world model. A first step is the visual detection of the obstacles in the image acquired by the robot. This is done using an algorithm based on radial search lines and colour-based blobs detection, where each obstacle is identified and delimited. After having the visually detected obstacles, a fusion with a-priori known information about the obstacles characteristics allows the obstacle separation and filtering, so that obstacles that don’t fill the criteria are discarded. With the position information shared by team mates, the matching of the obstacles and the team mates positions is also possible, thus identifying each of them. Finally, and with the purpose of having a team world model as coherent as possible, the robots are able to share the obstacle information of each other. The work presented in this paper was developed for the CAMBADA robotic soccer team. After achieving the 1st place in the Portuguese robotics open Robótica2008 and in the Robocup2008 world championship, the correct treatment of obstacles was one of the new challenges proposed among the team to improve the performance for the next competitions.

In social network analysis, a k-clique is a relaxed clique, i.e., a k-clique is a quasi-complete sub-graph. A k-clique in a graph is a sub-graph where the distance between any two vertices is no greater than k. The visualization of a small number of vertices can be easily performed in a graph. However, when the number of vertices and edges increases the visualization becomes incomprehensible. In this paper, we propose a new graph mining approach based on k-cliques. The concept of relaxed clique is extended to the whole graph, to achieve a general view, by covering the network with k-cliques. The sequence of k-clique covers is presented, combining small world concepts with community structure components. Computational results and examples are presented.

Financial distress prediction is of crucial importance in credit risk analysis with the increasing competition and complexity of credit industry. Although a variety of methods have been applied in this field, there are still some problems remained. The accurate and sensitive prediction in presence of unequal misclassification costs is an important one. Learning vector quantization (LVQ) is a powerful tool to solve financial distress prediction problem as a classification task. In this paper, a cost-sensitive version of LVQ is proposed which incorporates the cost information in the model. Experiments on two real data sets show the proposed approach is effective to improve the predictive capability in cost-sensitive situation.

This paper introduces an intelligent system that performs alarm correlation and root cause analysis. The system is designed to operate in large-scale heterogeneous networks from telecommunications operators. The proposed architecture includes a rules management module that is based in data mining (to generate the rules) and reinforcement learning (to improve rule selection) algorithms. In this work, we focus on the design and development of the rule generation part and test it using a large real-world dataset containing alarms from a Portuguese telecommunications company. The correlation engine achieved promising results, measured by a compression rate of 70% and assessed in real-time by experienced network administrator staff.

The identification of ”actionable” information in news stories has become a popular area for investigation. News presents some unique challenges for the researcher. The size constraints of a news story often require that full background information is omitted. Although this is acceptable for a human reader, it makes any form of automatic analysis difficult. Computational analysis may require some background information to provide context to news stories. There have been some attempts to identify and store background information. These approaches have tended to use an ontology to represent relationships and concepts present in the background information. The current methods of creating and populating ontologies with background information for news analysis were unsuitable for our future needs.

In this paper we present an automatic construction and population method of a domain ontology. This method produces an ontology which has the coverage of a manually created ontology and the ease of construction of the semi-automatic method. The proposed method uses a recursive algorithm which identifies relevant news stories from a corpus. For each story the algorithm tries to locate further related stories and background information. The proposed method also describes a pruning procedure which removes extraneous information from the ontology. Finally, the proposed method describes a procedure for adapting the ontology over time in response to changes in the monitored domain.

Collecting and annotating exemplary cases is a costly and critical task that is required in early stages of any classification process. Reducing labeling cost without degrading accuracy calls for a compromise solution which may be achieved with active learning. Common active learning approaches focus on accuracy and assume the availability of a pre-labeled set of exemplary cases covering all classes to learn. This assumption does not necessarily hold. In this paper we study the capabilities of a new active learning approach, d-Confidence, in rapidly covering the case space when compared to the traditional active learning confidence criterion, when the representativeness assumption is not met. Experimental results also show that d-Confidence reduces the number of queries required to achieve complete class coverage and tends to improve or maintain classification error.

This work address data stream mining from dynamic environments where the distribution underlying the observations may change over time. In these contexts, learning algorithms must be equipped with change detection mechanisms. Several methods have been proposed able to detect and react to concept drift. When a drift is signaled, most of the approaches use a forgetting mechanism, by releasing the current model, and start learning a new decision model, Nevertheless, it is not rare for the concepts from history to reappear, for example seasonal changes. In this work we present method that memorizes learnt decision models whenever a concept drift is signaled. The system uses meta-learning techniques that characterize the domain of applicability of previous learnt models. The meta-learner can detect re-occurrence of contexts and take pro-active actions by activating previous learnt models. The main benefit of this approach is that the proposed meta-learner is capable of selecting similar historical concept, if there is one, without the knowledge of true classes of examples.

This paper describes a data mining approach to the problem of detecting erroneous foreign trade transactions in data collected by the Portuguese Institute of Statistics (INE). Erroneous transactions are a minority, but still they have an important impact on the official statistics produced by INE. Detecting these rare errors is a manual, time-consuming task, which is constrained by a limited amount of available resources (e.g. financial, human). These constraints are common to many other data analysis problems (e.g. fraud detection). Our previous work addresses this issue by producing a ranking of outlyingness that allows a better management of the available resources by allocating them to the most relevant cases. It is based on an adaptation of hierarchical clustering methods for outlier detection. However, the method cannot be applied to articles with a small number of transactions. In this paper, we complement the previous approach with some standard statistical methods for outlier detection for handling articles with few transactions. Our experiments clearly show its advantages in terms of the criteria outlined by INE for considering any method applicable to this business problem. The generality of the approach remains to be tested in other problems which share the same constraints (e.g. fraud detection).

The basic concept of agent-based modeling is to create adaptive agents to operate in a changing environment. Agents make autonomous decisions and modify their environment through continuous interactions. The Functional Agent-Based Language for Simulations (FABLES) is a special purpose language for ABM that is intended to reduce programming skills required to create simulations. The aim of FABLES is to allow modelers to focus on modeling, and not on programming. This paper provides an overview of FABLES, explaining the traits and the design concepts of this hybrid language that merges features of object-oriented, functional and procedural languages to provide flexibility in model design. To demonstrate some of these issues, we describe modeling with FABLES via the popular El Farol Bar problem from a user perspective, by means of example.

The Airline Operations Control Centre (AOCC) tries to solve unexpected problems during the airline operation. Problems with aircraft, crewmembers and passengers are common and very hard to solve due to the several variables involved. This paper presents the implementation of a real-world multi-agent system for operations recovery in an airline company. The analysis and design of the system was done following a GAIA based methodology. We present the system specification as well as the implementation using JADE. A case study is included, where we present how the system solved a real problem.

Ecological models may be very complex due to the large number of physical, chemical, biological processes and variables and their interactions, leading to long simulation times. These models may be used to analyse different management scenarios providing support to decision-makers. Thus, the simultaneous simulation of different scenarios can make the process of analysis and decision quicker, provided that there are mechanisms to accelerate the generation of new scenarios and optimization of the choices between the results presented. This paper presents a new simulation platform – EcoSimNet – specially designed for environmental simulations, which allows the inclusion of intelligent agents and the introduction of parallel simulators to speed up and optimize the decision-making processes. Experiments were performed using EcoSimNet computational platform with an agent controlling several simulators and implementing a parallel version of the simulated annealing algorithm for optimizing aquaculture production. These experiments showed the capabilities of the framework, enabling a fast optimization process and making this work a step forward towards an agent based decision support system to optimize complex environmental problems.

Diplomacy is a 7-player game that requires coordination between players in order to achieve victory. Its huge search space makes existing search algorithms useless. In this paper we present Darkblade, a player designed as a Multi-Agent System that uses potential fields to calculate moves and evaluate board positions. We tested our player against other recent players. Although there are some limitations, the results are promising.

We provide an agent the capability to infer the relations (assertions) entailed by the rules that describe the formal semantics of an RDFS knowledge-base. The proposed inferencing process formulates each semantic restriction as a rule implemented within a SPARQL query statement. The process expands the original RDF graph into a fuller graph that explicitly captures the rule’s described semantics. The approach is currently being explored in order to support descriptions that follow the generic Semantic Web Rule Language. An experiment, using the Fire-Brigade domain, a small-scale knowledge-base, is adopted to illustrate the agent modeling method and the inferencing process.

One way of coordinating actions is by the adoption of norms: social conventions and lexicons are good examples of coordinating systems. This paper deals with the efficiency of the emergence of norms (adopted from a given initial set), inside a population of artificial agents that interact in pairs. Agents interact according to some well defined behavior each one implements. In order to conduct our work, we used a bench-mark agent behavior: the external majority, where agents keep a memory of its latest interactions, adopting the most observed choice occurring in the last m interactions, where m (memory size) is a given parameter. We present an empirical study in which we determine the best choices regarding the memory size that should be made in order to guarantee an efficient uniform decision emergence. In this context, a more efficient choice is one that leads to a smaller number of needed pair wise interactions. We performed a series of experiments with population sizes ranging from 50 to 5,000, memory ranging from 2 to 10, and for five network topologies (fully connected, regular, random, scale-free and small-world). Besides we also analyzed the impact on consensus emergence efficiency of the number of available initial choices (from 2, to the number of agents) together with different memory sizes.

Computational Trust and Reputation (CTR) systems are platforms capable of collecting trust information about candidate partners and of computing confidence scores for each one of these partners. These systems start to be viewed as vital elements in environments of electronic institutions, as they support fundamental decision making processes, such as the selection of business partners and the automatic and adaptive creation of contractual terms and associated enforcement methodologies. In this article, we propose a model for the aggregation of trust evidences that computes confidence scores taking into account dynamic properties of trust. We compare our model with a traditional statistical model that uses weighted means to compute trust, and show experimental results that show that in certain scenarios the consideration of the trust dynamics allows for a better estimation of confidence scores.

By using differential equations, evolutionary game theory shows that most of the games of competition for resources have equilibrium strategies named Evolutionary Stable. Although this approach can deduce these points, it is not possible to say how or whether a population will reach such equilibrium. We present an evolutionary agent-based model where individuals compete for space using mixed strategies. Agents belong to spatial locations that settle with whom they can interact, but they can freely move to contiguous partitions according to a definition of satisfiability. The simulation results show that, although the agents do not have any knowledge about equilibrium points, the population’s mean strategy always converges to a stable state, close and above to the analytic equilibrium. Moreover, it is reached independently of the initial population.

In social life, actors engage simultaneously in several relations with each other. The complex network of social links in which agents are engaged is fundamental for any realistic simulation of social life. Moreover, not only the existence of multiple-modality paths between agents in a simulation, but also the knowledge that those agents have about the quality and specificity of those links are relevant for the decisions the agents make and the consequences they have both at an individual and at a collective level. Each actor has a context in each of the relations that are represented as support of a simulation. We build on previous work about permeability between those contexts to study the novel notion of context switching. By switching contexts, individuals carry with them their whole set of personal characteristics to a different relation, while abandoning the previous one. When returning to the original context, all previous links are resumed. We apply these notions to a simple game: the consensus game, in which agents try to collectively achieve an arbitrary consensus through simple locally informed peer-to-peer interactions. We compare the results for context switching with results from simulating the simple game and the game with context permeability.

Normative environments for multi-agent systems provide means to monitor and enforce agents’ compliance to their commitments. However, when the normative space is imperfect, contracts to which norms apply may be unbalanced, and agents may exploit potential flaws to their own advantage. In this paper we analyze how a normative framework endowed with a simple adaptive deterrence sanctioning model responds to different agent populations. Agents are characterized by their risk tolerance and by their social attitude. We show that risk-averse or socially concerned populations cause lesser deterrence sanctions to be imposed by the normative system.

Risk, audit frequency, expected utility, decision on the rate of tax evasion: probably these words occur to the reader first about tax evasion modeling. However, it can easily turn out in the real world that the ’everyday evader’ hasn’t got reliable information about the risks of evasion, about the possible amount of fine, or about the efficiency of the tax authorities. The TAXSIM agent-based tax evasion model was developed to understand the macro-level taxpayer behavior better with its help. The model and first simulation results were presented on the ESSA 2008 conference. The aim of this article is to present a sensitivity analysis of the model. We applied Design of Experiments method to reveal the main parameter-response correlations on a selected parameter domain and used two extreme parameter sets to examine on what level the contradictory factors can compensate each other.

In this paper, we will address term translation extraction from indexed aligned parallel corpora, by using a couple of association measures combined by a voting scheme, for scaling down translation pairs according to the degree of internal cohesiveness, and evaluate results obtained. Precision obtained is clearly much better than results obtained in related work for the very low range of occurrences we have dealt with, and compares with the best results obtained in word translation.

This paper presents an approach for categorizing documents according to their implicit locational relevance. We report a thorough evaluation of several classifiers designed for this task, built by using support vector machines with multiple alternatives for feature vectors. Experimental results show that using feature vectors that combine document terms and URL n-grams, with simple features related to the locality of the document (e.g. total count of place references) leads to high accuracy values. The paper also discusses how the proposed categorization approach can be used to help improve tasks such as document retrieval or online contextual advertisement.

In order to automatically identify noun synonyms, we propose a new idea which opposes classical polysemous representations of words to monosemous representations based on the “

one sense per discourse

” hypothesis. For that purpose, we apply the attributional similarity paradigm on two levels: corpus and document. We evaluate our methodology on well-known standard multiple choice synonymy question tests and evidence that it steadily outperforms the baseline.

In this paper we consider the problem of building models that have high sentiment classification accuracy without the aid of a labeled dataset from the target domain. For that purpose, we present and evaluate a novel method based on level of abstraction of nouns. By comparing high-level features (e.g. level of affective words, level of abstraction of nouns) and low-level features (e.g. unigrams, bigrams), we show that, high-level features are better to learn subjective language across domains. Our experimental results present accuracy levels across domains of 71.2% using SVMs learning models.

In this paper, we will analyze the behavior of several parameters, namely type of contexts, similarity measures, and word space models, in the task of word similarity extraction from large corpora. The main objective of the paper will be to describe experiments comparing different extraction systems based on all possible combinations of these parameters. Special attention will be paid to the comparison between syntax-based contexts and windowing techniques, binary similarity metrics and more elaborate coefficients, as well as baseline word space models and Singular Value Decomposition strategies. The evaluation leads us to conclude that the combination of syntax-based contexts, binary similarity metrics, and a baseline word space model makes the extraction much more precise than other combinations with more elaborate metrics and complex models.

People are often asked to associate keywords to documents to enable applications to access the summarized core content of documents. This fact was the main motivation to work on an approach that may contribute to move from this manual procedure to an automatic one. Since Relevant Expressions (REs) or multi-word term expressions can be automatically extracted using the LocalMaxs algorithm, the most relevant ones can be used to describe the core content of each document. In this paper we present a language-independent approach for automatic generation of document descriptors. Results are shown for three different European languages and comparisons are made concerning different metrics for selecting the most informative REs of each document.

Though polarity classification has been extensively explored at various text levels and domains, there has been only comparatively little work looking into topic-related polarity classification. This paper takes a detailed look at how sentences expressing a polar attitude towards a given topic can be retrieved from a blog collection. A cascade of independent text classifiers on top of a sentence-retrieval engine is a solution with limited effectiveness. We show that more sophisticated processing is necessary. In this context, we not only investigate the impact of a more precise detection and disambiguation of polar expressions beyond simple text classification but also inspect the usefulness of a joint analysis of topic terms and polar expressions. In particular, we examine whether any syntactic information is beneficial in this classification task.

In this paper we propose an incremental item-based collaborative filtering algorithm. It works with binary ratings (sometimes also called implicit ratings), as it is typically the case in a Web environment. Our method is capable of incorporating new information in parallel with performing recommendation. New sessions and new users are used to update the similarity matrix as they appear. The proposed algorithm is compared with a non-incremental one, as well as with an incremental user-based approach, based on an existing explicit rating recommender. The use of techniques for working with sparse matrices on these algorithms is also evaluated. All versions, implemented in R, are evaluated on 5 datasets with various number of users and/or items. We observed that: Recall tends to improve when we continuously add information to the recommender model; the time spent for recommendation does not degrade; the time for updating the similarity matrix (necessary to the recommendation) is relatively low and motivates the use of the item-based incremental approach. Moreover we study how the number of items and users affects the user based and the item based approaches.