Carlos R. Lopes

A multiagent architecture for distance education systems

We describe a multiagent architecture for Web-based distance education systems that presents characteristics of intelligence and adaptability. This architecture is based on techniques of distributed artificial intelligence, planning and intelligent tutoring systems. As a result, the system allows adaptation of a given course to different types of students.

Automatic student modeling in adaptive educational systems through probabilistic learning style combinations: a qualitative comparison between two innovative stochastic approaches

Considering learning and how to improve students’ performances, adaptive educational systems must know the way in which an individual student learns best. In this context, this work presents a comparison between two innovative approaches to automatically detect and precisely adjust students’ learning styles during an adaptive course. These approaches take into account the nondeterministic and nonstationary aspects of learning styles. They are based upon two stochastic techniques: Markov chains and genetic algorithms. We found that the genetic algorithm (GA) based approach detects learning styles earlier and consequently provides personalized content earlier, making the learning process easier. The Markov based approach produces more fine-tuned results, taking into account strengths of learning styles.

Rule-Evolver: An Evolutionary Approach for Data Mining

This paper presents a genetic model and a software tool (Rule-Evolver) for the classification of records in Databases (DB). The model is based on the evolution of association rules of the IF-THEN type, which provide a high level of accuracy and coverage. The modeling of the Genetic Algorithm consists of the definition of chromosomes representation, the evaluation function, and the genetic operators. The Rule-Evolver is a tool that provides an environment for the evaluation of the genetic model and implements the interface with DBs. The case studies evaluate the performance of the model in several benchmark DBs. The results obtained are compared with those of other models, such as Artificial Neural Nets, Neuro-Fuzzy Systems and Statistical Models.

Detecção e Correção Automática de Estilos de Aprendizagem em Sistemas Adaptativos para Educação

Um dos aspectos mais importantes em sistemas adaptativos para educacaoe a capacidade de prover personalizacao de acordo com as necessidades especificasde cada estudante. Neste contexto, este trabalho apresenta uma abordagem promissorapara deteccao e correcao automatica de estilos de aprendizagem (EA) baseadaem cadeias de Markov. A maioria dos trabalhos nesta area apresentam abordagenscomplexas e ineficientes em algum aspecto. Alem disto, a abordagem apresentadaneste trabalho tem como vantagem tornar possivel aos estudantes o desenvolvimentode novas capacidades cognitivas, sendo baseada na combinacao de estilos de aprendizagem(CEA) e na correcao dinâmica de possiveis inconsistencias no modelo do estudante(ME), levando em consideracao o forte aspecto nao-deterministico do processode ensino-aprendizagem. Resultados promissores foram obtidos nos testes realizadoscom esta abordagem e sao discutidos neste trabalho.

Planning for resource production in real-time strategy games based on partial order planning, search and learning

Generally, a real-time strategy game is characterized by two stages. Initially, it is necessary to collect and produce resources. The next step is related to battles, taking into account the resources that were collected. The resources production stage is a key factor for winning the game. In this study the authors propose a mechanism for producing resources based on planning, supported by artificial intelligence using means-end analysis and scheduling. Emphasis is given to scheduling that uses an algorithm of real-time search and learning. The results show that the proposed system presents a better performance compared to related approaches.

Using Genetic Algorithms to minimize the distance and balance the routes for the multiple Traveling Salesman Problem

The Traveling Salesman Problem (TSP) has been used to model many real world applications. In this problem, a salesman travels using the shortest route between the cities that he must visit and returns to the depot. If it is required to use more than one salesman in the solution, the problem is called multiple Traveling Salesman Problem (mTSP), and the objective is to minimize the overall distance traveled by them. However, when only the overall distance is minimized, depending on the distribution of the cities, some salesmen tend to travel much more than others. This paper presents the development of Genetic Algorithms (GAs) to reduce both the overall distance and the difference between the distance traveled by each salesman. Since there are more than one objective to be optimized, two approaches were evaluated. A multi-objective GA and a mono-objective GA with a fitness function that combines both objectives. In order to find the best results for each approach, different methods for crossover and selection have been used. Experimental results show the effectiveness of the proposed GAs. The results further indicate that, considering both objectives, the multi-objective GA generates more balanced solutions for almost all instances.

Maximization of the resource production in RTS games through stochastic search and planning

RTS games are an important field of research in Artificial Intelligence Planning. These games have many challenges for planning. RTS games are characterized by two important phases. The first one has to do with gathering resources and developing an army. In the second phase the resources produced are used in battles against enemies. Thus, the first phase is vital for success in the game and the power of the army developed directly reflects in the chances of victory. This work focuses on the choice of goals to be achieved during the game. To do this, we developed an approach for maximization of production resources based on stochastic search and planning. The results show the effectiveness of our approach in finding goals that increase the strength of the player army.

Obstacle avoidance for mobile robots: A Hybrid Intelligent System based on Fuzzy Logic and Artificial Neural Network

Obstacle avoidance is one of the most important aspects of autonomous mobile robots. This task is composed by two phases. First, the robot must detect obstacles in the environment with its sensors. Then, it must choose an appropriate movement to go through the environment without colliding. However, the noise produced during the sensors reading can lead the robot to take wrong decisions. This paper presents the development of an E-Puck mobile robot obstacle avoidance controller using a Hybrid Intelligent System (HIS) based on Fuzzy Logic (FL) and Artificial Neural Networks (ANN). The FL treats the data of infrared sensors and then feeds an ANN that decides which movement the robot must perform. The HIS was compared to another approach in which the data of infrared sensors are used directly in an ANN. The empirical results show that HIS avoids more collisions and improves the smoothness of navigation.

Probabilistic Planning for Multiple Stocks of Financial Markets

In the financial market the decision on when buying or selling stocks is fundamental in order to achieve profit. There are several techniques that can be used to help investors in order to make a decision. One of those is the employment of technical analysis that consists of chart studies concerning the behaviour of stock prices. In this paper we describe our approach for this problem of decision making, which is cast as a planning problem in the presence of uncertainties. We propose the use of Partially Observable Markov Decision Process (POMDP) for the task of planning the negotiation of stocks on the financial market. The main and desired contribution consist of exploring this type of planning using multiple stocks. The stocks are selected from correlation calculations. The use of multiple stocks provided better results, when compared to other researched strategies.

Generating Plans Using LRTA

A new category of faster and efficiently planners has been used to solve planning problems, which use heuristic search algorithms to build their plans. One of the first successful planning systems to use this new approach was the Fast Forward (FF) planner, in which many other recent planners have built upon. FF proposed a heuristic function for planning problems and created a strategy that combines the execution of the algorithms Enforced Hill Climbing (EHC) and Best First Search (BFS). Although this method presents an enhanced performance when compared to alternative methods, it has some weaknesses. In this paper we present a new planning method based on LRTA, which is an algorithm guided by heuristics like EHC and is complete as BFS. It has some optimizations like pruning successors during expansion and a heap with maximum capacity to store the states during the search. The authors also developed a test environment for planning algorithms. Experiments carried out in this environment show signicant results when compared to FF.