Catálogo de publicaciones - libros

Nearest Prototype methods can be quite successful on many pattern classification problems. In these methods, a collection of prototypes has to be found that accurately represents the input patterns. The classifier then assigns classes based on the nearest prototype in this collection. In this paper we develop a new algorithm (called AMPSO), based on the Particle Swarm Optimization (PSO) algorithm, that can be used to find those prototypes. Each particle in a swarm represents a single prototype in the solution; the swarm evolves using modified PSO equations with both particle competition and cooperation. Experimentation includes an artificial problem and six common application problems from the UCI data sets. The results show that the AMPSO algorithm is able to find solutions with a reduced number of prototypes that classify data with comparable or better accuracy than the 1-NN classifier. The algorithm can also be compared or improves the results of many classical algorithms in each of those problems; and the results show that AMPSO also performs significantly better than any tested algorithm in one of the problems.

This paper presents a new framework for behavioural modelling that allows to unravel the key drivers that direct specific cognitive behaviours. In order to do so, a novel framework for clustering based on utility measures is presented that allows to understand the different behaviours that different groups of people may have, and allows the creation of profiles that are relevant with respect to the utility measure. The proposed method is not contrary to other clustering methods but rather builds on the functionality of ’basic’ clustering algorithms. A common aim of clustering consists of partitioning a set of patterns into different subsets of patterns which have homogeneous characteristics. In this paper we suggest a more ambitious goal that additionally tries to maximize a utility measure. The paper also describes the results obtained when the method is used to analyze human behaviour in the area of customer intelligence. Specifically, the paper analyzes human behaviour with respect to different socio-demographic and economic indicators and allows to uncover the underlying characteristics that may explain the observed cognitive behaviour.

In this paper, a multilevel approach of Ant Colony Optimization to solve the Traveling Salesman Problem is introduced. The basic idea is to split the heuristic search performed by ants into two stages; in this case we use both the Ant System and Ant Colony System algorithms. Also, the effect of using local search was analyzed. We have studied the performance of this new algorithm for several Traveling Salesman Problem instances. Experimental results obtained conclude that the Two-Stage approach significantly improves the Ant System and Ant Colony System in terms of the computation time needed.

This paper is about Set Partitioning formulation and resolution for a particular case of VRP, the Dial-a-ride Problem. Set Partitioning has demonstrated to be useful modeling this problem and others very visible and economically significant problems. But the main disadvantage of this model is the need to explicitly generate a large set of possibilities to obtain good solutions. Additionally, in many cases a prohibitive time is needed to find the exact solution. Nowadays, many efficient metaheuristic methods have been developed to make possible a good solution in a reasonable amount of time. In this work we try to solve it with Low-level Hybridizations of Ant Colony Optimization and Constraint Programming techniques helping the construction phase of the ants. Computational results solving some benchmark instances are presented showing the advantages of using this kind of hybridization.

This paper employs a binary discrete version of the classical Particle Swarm Optimization to compare the maximum net present value achieved by a gas turbines biomass plant and a gas engine biomass plant. The proposed algorithm determines the optimal location for biomass turbines plant and biomass gas engine plant in order to choose the most profitable between them. Forest residues are converted into biogas . The fitness function for the binary optimization algorithm is the net present value. The problem constraints are: the generation system must be located inside the supply area, and its maximum electric power is 5 MW. Computer simulations have been performed using 20 particles in the swarm and 50 iterations for each kind of power plant. Simulation results indicate that Particle Swarm Optimization is a useful tool to choose successful among different types of biomass plant technologies. In addition, the comparison is made with reduced computation time (more than 800 times lower than that required for exhaustive search).

In this paper a hybrid approach to the autonomous navigation of robots in cluttered environment with unknown obstacles is introduced. It is shown the efficiency of the hybrid solution by combining the optimization power of evolutionary algorithms and at the same time the efficiency of the Reinforcement Learning in real-time and on-line situations. Experimental results concerning the navigation of a L-shaped robot in a cluttered environment with unknown obstacles in which appear real-time and on-line constraints well-suited to RL algorithms and extremely high dimension of the state space usually unpractical for RL algorithms but at the same time well-suited to evolutionary algorithms, are also presented. The experimental results confirm the validity of the hybrid approach to solve hard real-time, on-line and high dimensional robot motion control problems.

Reflexes are meant to provide animals with automatic responses for a better adaptation to their niches. In particular, humans have the capability to voluntarily modify these responses in certain situations to attain specific goals. The ability of using past experiences to tune automatic responses (reflexes) has contributed to a better adaptation to our environments and thus, the question arises of applying this to machines. In the robotic arena, imitating animal reflexes has been largely explored through fixed stimuli-behavior schemas included in reactive or hybrid architectures. In this paper we consider the less explored direction of permitting a mobile robot to modify its reflexes according to its experience, i.e. ignoring the reflex of stopping when approaching an obstacle if the robot goal is close. We explore reinforcement learning as a mechanism to automatically learn when and how modulate reflexes over the robot operational life. Advantages of our mechanism are illustrated in simulations.

We investigate how it is possible to shape robot behaviour adopting a molecular or molar point of view. These two ways to approach the issue are inspired by Learning Psychology, whose famous representatives suggest different ways of intervening on animal behaviour.

Starting from this inspiration, we apply these two solutions to Evolutionary Robotics’ models. Two populations of simulated robots, controlled by Artificial Neural Networks are evolved using Genetic Algorithms to wander in a rectangular enclosure. The first population is selected by measuring the wandering behaviour at micro-actions level, the second one is evaluated by considering the macro-actions level. Some robots are evolved with a molecular fitness function, while some others with a molar fitness function. At the end of the evolutionary process, we evaluate both populations of robots on behavioral, evolutionary and latent-learning parameters.

Choosing what kind of behaviour measurement must be employed in an evolutionary run depends on several factors, but we underline that a choice that is based on self-organization, emergence and autonomous behaviour principles, the basis Evolutionary Robotics lies on, is perfectly in line with a molar fitness function.

Isotropic Sequence Order learning (ISO-learning) and Input Correlation Only learning (ICO-learning) are unsupervised neural algorithms to learn temporal differences. The use of devices implementing this algorithms by simulation in reactive neural networks is proposed. We have applied several modifications to original rules: weights sign restriction, to adequate ISO-learning and ICO-learning devices outputs to the usually predefined kinds of connections (excitatory/inhibitory) used in neural networks, and decay term inclusion for weights stabilization. Original experiments with these algorithms are replicated as accurate as possible with a simulated robot and a discretization of the algorithms. Results are similar to those obtained in original experiments with analogue devices.

The context of this paper is the auto calibration of a CCD low cost camera of a robotic pet. The underlined idea of the auto calibration is to imitate the human eye capabilities, which is able to accommodates changing lighting conditions, and only when all functionalities works properly, light is converted to impulses to the brain where the image is sensed. In order to choose the more appropriated camera’s parameters, a fuzzy rules model has been generated following a neuro-fuzzy approach. This model classifies images into five classes: from very dark, to very light. This is the first step to the generation of a subsequent fuzzy controller able to change the camera setting in order to improve the image received from an environment with changing lighting conditions.