http://sedici.unlp.edu.ar:80/handle/10915/182956 Thu, 12 Nov 2026 17:25:05 GMT 2026-11-12T17:25:05Z http://sedici.unlp.edu.ar:80/handle/10915/185599 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Raising goats is a strong component of regional economies and a main activity of little farmers in Argentina. Knowing the diet is fundamental to correct nutritional lacks resulting from grazing native species, as it will help to design management strategies both for the native forage resource and for the nutritional state of the animal. This fact has a preponderant relevance when goat production is done at natural areas. Microhistology is one of the more frequently used methodology to determine the botanical composition of herbivorous diet. The specialist working at a microscope, according to his training and experience, can identify the species taken by the animal and to which those cells belong. Neural network approaches are suited for recognition problems, this article presents a backpropagation neural network initially trained to identify the four most significant vegetal species found in goat diet that graze freely in the Semiarid Chaco of North-West Argentina. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/185599 2002-01-01T00:00:00Z Raising goats is a strong component of regional economies and a main activity of little farmers in Argentina. Knowing the diet is fundamental to correct nutritional lacks resulting from grazing native species, as it will help to design management strategies both for the native forage resource and for the nutritional state of the animal. This fact has a preponderant relevance when goat production is done at natural areas. Microhistology is one of the more frequently used methodology to determine the botanical composition of herbivorous diet. The specialist working at a microscope, according to his training and experience, can identify the species taken by the animal and to which those cells belong. Neural network approaches are suited for recognition problems, this article presents a backpropagation neural network initially trained to identify the four most significant vegetal species found in goat diet that graze freely in the Semiarid Chaco of North-West Argentina. http://sedici.unlp.edu.ar:80/handle/10915/185598 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) One of the main concerns in the simulation of adaptive behaviors is the study of the complexity of the systems. These studies are carried out through the construction of bottom-up, decentralized models. In this work, we present a decentralized model to study the Vertical Cultural Transmission (VCT), i.e., information transmission from parents to children, as a catalytic factor over the interaction between individual learning and population evolution. Our model is based on the incorporation of advice, in the form of suggestions, as a way of information transmission between agents. An evolutionary computational technique called classifier systems is used as the mechanism for individual learning. In particular, a recent variant based on the accuracy of the prediction made by the rules, called XCS, is applied. In order to control the environmental complexity we adopt the Latent Energy Environment model (LEE). Finally, experimental results showing that vertical cultural transmission does not improve the adaptation of the populations to the described environment are presented. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/185598 2002-01-01T00:00:00Z One of the main concerns in the simulation of adaptive behaviors is the study of the complexity of the systems. These studies are carried out through the construction of bottom-up, decentralized models. In this work, we present a decentralized model to study the Vertical Cultural Transmission (VCT), i.e., information transmission from parents to children, as a catalytic factor over the interaction between individual learning and population evolution. Our model is based on the incorporation of advice, in the form of suggestions, as a way of information transmission between agents. An evolutionary computational technique called classifier systems is used as the mechanism for individual learning. In particular, a recent variant based on the accuracy of the prediction made by the rules, called XCS, is applied. In order to control the environmental complexity we adopt the Latent Energy Environment model (LEE). Finally, experimental results showing that vertical cultural transmission does not improve the adaptation of the populations to the described environment are presented. http://sedici.unlp.edu.ar:80/handle/10915/183292 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) When we search for a solution of a problem employing an evolutionary computation, information of a geometry of fitness landscape is sometimes very helpful. What does the fitness landscape look like? How many global peaks are there in the fitness landscape? Do Local peaks exist? How does the shape of these peaks look? To study these issues, we propose a simple technique of learning a shape of a peak in the fitness landscape by observing a cross-section of the peak by a hyperplane. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183292 2002-01-01T00:00:00Z When we search for a solution of a problem employing an evolutionary computation, information of a geometry of fitness landscape is sometimes very helpful. What does the fitness landscape look like? How many global peaks are there in the fitness landscape? Do Local peaks exist? How does the shape of these peaks look? To study these issues, we propose a simple technique of learning a shape of a peak in the fitness landscape by observing a cross-section of the peak by a hyperplane. http://sedici.unlp.edu.ar:80/handle/10915/183289 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Many approaches to distributed load balancing focus on algorithms, most of the time leading to centralized mechanisms while trying to solve distributed problems. Olafsson's approach is interesting in that it offers a framework where local decision making is the key to achieving efficient resource allocation. Although he decouples interactions between agents, this loose coupling is lost with the introduction of differentiated pricing strategies and makes the stability analysis around potential equilibria more complex. Furthermore, his model does not provide support for stochastic events in distributed environments. In this paper, we describe a modi fication of his model to address extinction and innovation in the population of resources and present our current experimental directions. This paper is the first installment of a series which will describe the architecture of the system we have built to prove our concepts and provide experimental results. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183289 2002-01-01T00:00:00Z Many approaches to distributed load balancing focus on algorithms, most of the time leading to centralized mechanisms while trying to solve distributed problems. Olafsson's approach is interesting in that it offers a framework where local decision making is the key to achieving efficient resource allocation. Although he decouples interactions between agents, this loose coupling is lost with the introduction of differentiated pricing strategies and makes the stability analysis around potential equilibria more complex. Furthermore, his model does not provide support for stochastic events in distributed environments. In this paper, we describe a modi fication of his model to address extinction and innovation in the population of resources and present our current experimental directions. This paper is the first installment of a series which will describe the architecture of the system we have built to prove our concepts and provide experimental results. http://sedici.unlp.edu.ar:80/handle/10915/183286 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Many approaches to distributed load balancing focus on algorithms, most of the time leading to centralized mechanisms while trying to solve distributed problems. Olafsson's approach is interesting in that it offers a framework where local decision making is the key to achieving efficient resource allocation. Although he decouples interactions between agents, this loose coupling is lost with the introduction of differentiated pricing strategies and makes the stability analysis around potential equilibria more complex. Furthermore, his model does not provide support for stochastic events in distributed environments. In this paper, we describe a modi fication of his model to address extinction and innovation in the population of resources and present our current experimental directions. This paper is the first installment of a series which will describe the architecture of the system we have built to prove our concepts and provide experimental results. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183286 2002-01-01T00:00:00Z Many approaches to distributed load balancing focus on algorithms, most of the time leading to centralized mechanisms while trying to solve distributed problems. Olafsson's approach is interesting in that it offers a framework where local decision making is the key to achieving efficient resource allocation. Although he decouples interactions between agents, this loose coupling is lost with the introduction of differentiated pricing strategies and makes the stability analysis around potential equilibria more complex. Furthermore, his model does not provide support for stochastic events in distributed environments. In this paper, we describe a modi fication of his model to address extinction and innovation in the population of resources and present our current experimental directions. This paper is the first installment of a series which will describe the architecture of the system we have built to prove our concepts and provide experimental results. http://sedici.unlp.edu.ar:80/handle/10915/183284 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Current research on artificial vision and pattern recognition tends to concentrate either on numerical processing (filtering, morphological, spectral) or in symbolic or subsymbolic processing (neural networks, fuzzy logic, knowledge-based systems). In this work we combine both kinds of processing in a hybrid image processing architecture. The numerical processing part implements the most usual facilities (equalization, convolution filters, morphological filters, segmentation and description) in a way adequate to transform the input image into a polygonal outline. Then recognition is performed with a rule-based system implemented in Prolog. This allows a neat high-level representation of the patterns to recognize as a set of logical relations, and also the recognition procedure is represented as a set of logical rules. To integrate the numerical and logical components of our system, we embedded a Prolog interpreter as a software component within a visual programming language. Thus, our architecture features both the speed and versatility of a visual language application, and the abstraction level and modularity of a logical description. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183284 2002-01-01T00:00:00Z Current research on artificial vision and pattern recognition tends to concentrate either on numerical processing (filtering, morphological, spectral) or in symbolic or subsymbolic processing (neural networks, fuzzy logic, knowledge-based systems). In this work we combine both kinds of processing in a hybrid image processing architecture. The numerical processing part implements the most usual facilities (equalization, convolution filters, morphological filters, segmentation and description) in a way adequate to transform the input image into a polygonal outline. Then recognition is performed with a rule-based system implemented in Prolog. This allows a neat high-level representation of the patterns to recognize as a set of logical relations, and also the recognition procedure is represented as a set of logical rules. To integrate the numerical and logical components of our system, we embedded a Prolog interpreter as a software component within a visual programming language. Thus, our architecture features both the speed and versatility of a visual language application, and the abstraction level and modularity of a logical description. http://sedici.unlp.edu.ar:80/handle/10915/183233 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Automation is an important aspect of mass appraisal because of quality and time requirements. Regression analysis has been used in computer-assisted mass appraisal (CAMA) for a long time but suffer of problems, like functional form miss-specification and multicollinearity, since real estate markets potentially have nonlinear effects and inter-related variables. In the last years, several papers presented artificial neural networks (ANNs) like an alternative tool to regression, with good results mainly when large samples were used. But ANN models have problems in the explanation of the results, because of its “black box nature”. This is a serious problem in fields like taxation since the taxpayer needs to be informed about the way of assessment. However, there are methods able to explain ANNs results by extracting the knowledge stored in the network parameters. Fuzzy rules extracting is the most commonly referred method, which the rules are generated from network’s weights. This paper presents a case study comparing the results of neuro-fuzzy and regression models using data of real estate market of Porto Alegre, a Brazilian city, and discussing their advantages and disadvantages. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183233 2002-01-01T00:00:00Z Automation is an important aspect of mass appraisal because of quality and time requirements. Regression analysis has been used in computer-assisted mass appraisal (CAMA) for a long time but suffer of problems, like functional form miss-specification and multicollinearity, since real estate markets potentially have nonlinear effects and inter-related variables. In the last years, several papers presented artificial neural networks (ANNs) like an alternative tool to regression, with good results mainly when large samples were used. But ANN models have problems in the explanation of the results, because of its “black box nature”. This is a serious problem in fields like taxation since the taxpayer needs to be informed about the way of assessment. However, there are methods able to explain ANNs results by extracting the knowledge stored in the network parameters. Fuzzy rules extracting is the most commonly referred method, which the rules are generated from network’s weights. This paper presents a case study comparing the results of neuro-fuzzy and regression models using data of real estate market of Porto Alegre, a Brazilian city, and discussing their advantages and disadvantages. http://sedici.unlp.edu.ar:80/handle/10915/183232 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) En este trabajo se presenta el diseño de una red de sensores de costo mínimo sujeto a restricciones de estimabilidad y precisión sobre un conjunto de variables claves. Para la resolución del problema se utiliza un enfoque orientado a ecuaciones basado en un algoritmo evolutivo (AE) que incorpora conocimiento específico del problema. El algoritmo utiliza el concepto de poblaciones espacialmente estructuradas de manera tal de lograr un balance adecuado entre exploración y explotación del espacio de búsqueda. El método se aplica a un problema tomado de la literatura y se comparan los resultados obtenidos por enumeración explícita. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183232 2002-01-01T00:00:00Z En este trabajo se presenta el diseño de una red de sensores de costo mínimo sujeto a restricciones de estimabilidad y precisión sobre un conjunto de variables claves. Para la resolución del problema se utiliza un enfoque orientado a ecuaciones basado en un algoritmo evolutivo (AE) que incorpora conocimiento específico del problema. El algoritmo utiliza el concepto de poblaciones espacialmente estructuradas de manera tal de lograr un balance adecuado entre exploración y explotación del espacio de búsqueda. El método se aplica a un problema tomado de la literatura y se comparan los resultados obtenidos por enumeración explícita. http://sedici.unlp.edu.ar:80/handle/10915/183231 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) The objective of this work is the consideration and implementation of some basic premises used in the learning process in Artificial Neural Networks (ANN`s). Initially the net will be trained starting from the Neusciences simulator: ActiveX to, starting from the result of this simulation, be compared with the algorithm of competitive learning through the Kohonen Self-Organizable Map. The chosen domain for the implementation of the learning algorithms was the application in the Clinical Diagnosis of the Convulsive Crises, based on the International Classification League Against Epilepsy ILAI/81. According to the results of the simulator and using the Learning Vector Quantization (LVQ1) technique with the 2x2 configuration, the base of training of the network showed a performance of 69,76 and 71,31% respectively. For the test set of the simulator and the LVQ1 technique the network obtained an index satisfactory of classification of 80% and 100% respectively. With the 5x5 configuration to increase the index of classification. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183231 2002-01-01T00:00:00Z The objective of this work is the consideration and implementation of some basic premises used in the learning process in Artificial Neural Networks (ANN`s). Initially the net will be trained starting from the Neusciences simulator: ActiveX to, starting from the result of this simulation, be compared with the algorithm of competitive learning through the Kohonen Self-Organizable Map. The chosen domain for the implementation of the learning algorithms was the application in the Clinical Diagnosis of the Convulsive Crises, based on the International Classification League Against Epilepsy ILAI/81. According to the results of the simulator and using the Learning Vector Quantization (LVQ1) technique with the 2x2 configuration, the base of training of the network showed a performance of 69,76 and 71,31% respectively. For the test set of the simulator and the LVQ1 technique the network obtained an index satisfactory of classification of 80% and 100% respectively. With the 5x5 configuration to increase the index of classification. http://sedici.unlp.edu.ar:80/handle/10915/183226 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) We propose a simple method for the reconstruction of slow dynamics perturbations from non-stationary time series records. The method traces the evolution of the perturbing signal by simultaneously learning the intrinsic stationary dynamics and the time dependency of the changing parameter. For this purpose, an extra input unit is added to a feedforward artificial neural network and a suitable error function minimized in the training process. Testing of our algorithm on synthetic data shows its efficacy and allows extracting general criteria for applications on real-world problems. Finally, a preliminary study of the well-known sunspot time series recovers particular features of this series, including recently reported changes in solar activity during last century. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183226 2002-01-01T00:00:00Z We propose a simple method for the reconstruction of slow dynamics perturbations from non-stationary time series records. The method traces the evolution of the perturbing signal by simultaneously learning the intrinsic stationary dynamics and the time dependency of the changing parameter. For this purpose, an extra input unit is added to a feedforward artificial neural network and a suitable error function minimized in the training process. Testing of our algorithm on synthetic data shows its efficacy and allows extracting general criteria for applications on real-world problems. Finally, a preliminary study of the well-known sunspot time series recovers particular features of this series, including recently reported changes in solar activity during last century. http://sedici.unlp.edu.ar:80/handle/10915/183224 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) The Feature Subset Selection is an important problem within the Machine Learning area where the learning algorithm is faced with the problem of selecting relevant features while ignoring the rest. Another important problem within this area is the complexity of the knowledge acquired (hypotheses) though rules induction. Rough Sets Theory is a mathematical tool to deal with vagueness and uncertainty information. One of the main features of this approach are the reducts, which is a minimal feature set that preserves the ability to discern each object from the others. This work presents in detail several experiments, results and comparisons using Rough Sets Reducts and other Filters for feature subset selection and rule induction. The purpose of this work is to investigate the reduction of the complexity of the rules induced in terms of the Feature Subset Selection problem, considering as measure of rules complexity the number of rules induced. All the experiments where run on natural datasets, most of them obtained from the UCI Irvine Repository. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183224 2002-01-01T00:00:00Z The Feature Subset Selection is an important problem within the Machine Learning area where the learning algorithm is faced with the problem of selecting relevant features while ignoring the rest. Another important problem within this area is the complexity of the knowledge acquired (hypotheses) though rules induction. Rough Sets Theory is a mathematical tool to deal with vagueness and uncertainty information. One of the main features of this approach are the reducts, which is a minimal feature set that preserves the ability to discern each object from the others. This work presents in detail several experiments, results and comparisons using Rough Sets Reducts and other Filters for feature subset selection and rule induction. The purpose of this work is to investigate the reduction of the complexity of the rules induced in terms of the Feature Subset Selection problem, considering as measure of rules complexity the number of rules induced. All the experiments where run on natural datasets, most of them obtained from the UCI Irvine Repository. http://sedici.unlp.edu.ar:80/handle/10915/183223 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) The Situation Calculus has been used as a tool to develop planning systems. However, the traditional Situation Calculus does not supply mechanisms to represent actions that consume resources and represent time, which are features that are important in applications like Project Management. This work presents an extension of Situation Calculus which allows representation of time description and actions that consume human and material resources. The application of this extended calculus is exemplified by a project planning case of the telecommunications area. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183223 2002-01-01T00:00:00Z The Situation Calculus has been used as a tool to develop planning systems. However, the traditional Situation Calculus does not supply mechanisms to represent actions that consume resources and represent time, which are features that are important in applications like Project Management. This work presents an extension of Situation Calculus which allows representation of time description and actions that consume human and material resources. The application of this extended calculus is exemplified by a project planning case of the telecommunications area. http://sedici.unlp.edu.ar:80/handle/10915/183219 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Combinatorial optimization problems require combinatorial-specific search operators so that populations of candidate solutions can evolve efficiently. Indeed, several researchers created modifications to the basic genetic operators of mutation and recombination in order to create high performing combinatorial-specific operators. However, it is not known which operators perform better as no systematic comparisons have been done. In this work, a new algorithm that explores a new chromosomal organization based on multigene families is used. This new organization together with several combinatorialspecific search operators, namely, inversion, gene and sequence deletion/insertion, and restricted and generalized permutation, allow the algorithm to perform with high efficiency. The performance of the new algorithm is empirically compared on the 13- and 19-cities tour traveling salesperson problem, showing that the long abandoned inversion operator is by far the most efficient of the combinatorial operators. The efficiency and potentialities of the new algorithm are further demonstrated by solving a simple task assignment problem. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183219 2002-01-01T00:00:00Z Combinatorial optimization problems require combinatorial-specific search operators so that populations of candidate solutions can evolve efficiently. Indeed, several researchers created modifications to the basic genetic operators of mutation and recombination in order to create high performing combinatorial-specific operators. However, it is not known which operators perform better as no systematic comparisons have been done. In this work, a new algorithm that explores a new chromosomal organization based on multigene families is used. This new organization together with several combinatorialspecific search operators, namely, inversion, gene and sequence deletion/insertion, and restricted and generalized permutation, allow the algorithm to perform with high efficiency. The performance of the new algorithm is empirically compared on the 13- and 19-cities tour traveling salesperson problem, showing that the long abandoned inversion operator is by far the most efficient of the combinatorial operators. The efficiency and potentialities of the new algorithm are further demonstrated by solving a simple task assignment problem. http://sedici.unlp.edu.ar:80/handle/10915/183217 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) How to generate and aggregate base learners to have optimal ensemble generalization capabilities is an important questions in building composite regression/classification machines. We present here an evaluation of several algorithms for artificial neural networks aggregation in the regression settings, including new proposals and comparing them with standard methods in the literature. We also discuss a potential problem with sequential algorithms: the non-frequent but damaging selection through their heuristics of particularly bad ensemble members. We show that one can cope with this problem by allowing individual weighting of aggregate members. Our algorithms and their weighted modifications are favorably tested against other methods in the literature, producing a performance improvement on the standard statistical databases used as benchmarks. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183217 2002-01-01T00:00:00Z How to generate and aggregate base learners to have optimal ensemble generalization capabilities is an important questions in building composite regression/classification machines. We present here an evaluation of several algorithms for artificial neural networks aggregation in the regression settings, including new proposals and comparing them with standard methods in the literature. We also discuss a potential problem with sequential algorithms: the non-frequent but damaging selection through their heuristics of particularly bad ensemble members. We show that one can cope with this problem by allowing individual weighting of aggregate members. Our algorithms and their weighted modifications are favorably tested against other methods in the literature, producing a performance improvement on the standard statistical databases used as benchmarks. http://sedici.unlp.edu.ar:80/handle/10915/183215 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) An object-oriented framework to support the modeling and management of the design process (DPR) is introduced. It naturally integrates the representation of both the DPR itself, and the outcomes that are achieved as the result of the various design activities. The integral view of tracing adopted not only captures and manages the products being generated but also the activities that occurred, their associated context and the adopted decisions. The Version Administration System introduced in this paper provides an explicit mechanism to manage the different model versions being generated during the course of a design project as design activities are executed. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183215 2002-01-01T00:00:00Z An object-oriented framework to support the modeling and management of the design process (DPR) is introduced. It naturally integrates the representation of both the DPR itself, and the outcomes that are achieved as the result of the various design activities. The integral view of tracing adopted not only captures and manages the products being generated but also the activities that occurred, their associated context and the adopted decisions. The Version Administration System introduced in this paper provides an explicit mechanism to manage the different model versions being generated during the course of a design project as design activities are executed. http://sedici.unlp.edu.ar:80/handle/10915/183169 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) In this paper we discuss the problem of reinforcement learning in one environment and applying the policy obtained to other environments. We first state a method to evaluate the utility of a policy. We then propose a general model to apply one policy to different environments and compare them. To illustrate the theory we present examples for an obstacle avoidance behavior in various block world environments. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183169 2002-01-01T00:00:00Z In this paper we discuss the problem of reinforcement learning in one environment and applying the policy obtained to other environments. We first state a method to evaluate the utility of a policy. We then propose a general model to apply one policy to different environments and compare them. To illustrate the theory we present examples for an obstacle avoidance behavior in various block world environments. http://sedici.unlp.edu.ar:80/handle/10915/183168 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) In this paper, we discuss the role and utility of synthetic vision in computer games. We present an implementation of a synthetic vision module based on two viewports rendered in real-time, one representing static information and the other dynamic, with false colouring being used for object identification, depth information and movement representation. We demonstrate the utility of this synthetic vision module by using it as input to a simple rule-based AI module that controls agent behaviour in a first-person shooter game. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183168 2002-01-01T00:00:00Z In this paper, we discuss the role and utility of synthetic vision in computer games. We present an implementation of a synthetic vision module based on two viewports rendered in real-time, one representing static information and the other dynamic, with false colouring being used for object identification, depth information and movement representation. We demonstrate the utility of this synthetic vision module by using it as input to a simple rule-based AI module that controls agent behaviour in a first-person shooter game. http://sedici.unlp.edu.ar:80/handle/10915/183167 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) In this paper, we study the problem of identification of underlying causes from a mixture of causes when these causes interfere with one another in a particular manner: the causes we consider are visual inputs to a neural network system which are created by independent underlying causes which may occlude each other. The prototypical problem in this area is a mixture of horizontal and vertical bars in which each horizontal bar interferes with the representation of each vertical bar and vice versa. We develop an unsupervised neural network which is able to identify a single orientation of bars from a mixture of horizontal and vertical bars. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183167 2002-01-01T00:00:00Z In this paper, we study the problem of identification of underlying causes from a mixture of causes when these causes interfere with one another in a particular manner: the causes we consider are visual inputs to a neural network system which are created by independent underlying causes which may occlude each other. The prototypical problem in this area is a mixture of horizontal and vertical bars in which each horizontal bar interferes with the representation of each vertical bar and vice versa. We develop an unsupervised neural network which is able to identify a single orientation of bars from a mixture of horizontal and vertical bars. http://sedici.unlp.edu.ar:80/handle/10915/183166 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) This work firstly presents an approach based on the organizational concept of business processes to identify roles and protocols as part of the analysis stage of a methodology for agent-oriented analysis and design. Secondly, the work presents an approach to help to decide when using a mobile agent in a multi-agent architecture is better than using a static one. This approach is based on both the quality attributes specified for the multi-agent architecture and the execution environments of the multi-agent system. A real case is used to exemplify these approaches. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183166 2002-01-01T00:00:00Z This work firstly presents an approach based on the organizational concept of business processes to identify roles and protocols as part of the analysis stage of a methodology for agent-oriented analysis and design. Secondly, the work presents an approach to help to decide when using a mobile agent in a multi-agent architecture is better than using a static one. This approach is based on both the quality attributes specified for the multi-agent architecture and the execution environments of the multi-agent system. A real case is used to exemplify these approaches. http://sedici.unlp.edu.ar:80/handle/10915/183165 Objeto de conferencia Simposio Argentino de Inteligencia Artificial (ASAI 2002) - JAIIO 31 (Santa Fe, 9 al 13 de septiembre de 2002) Nowadays, Java-based platforms are the most common proposals for building mobile agent systems using web technology. However, the weak mobility model they use, the lack of adequate support for developing inference and reasoning mechanisms, added to the inherent complexity of building location aware software, impose strong limitations for developing mobile intelligent agent systems. In this article we present MoviLog, a platform for building Prolog-based mobile agents with a strong mobility model. MoviLog is an extension of JavaLog, an integration of Java and Prolog, that allows a user to take advantage of the best features of the programming paradigms they represent. MoviLog provides logic modules, called Brainlets, which are able to migrate among different web sites, either proactively or reactively, to use the available knowledge in order to find a solution. The most interesting feature introduced by MoviLog is the reactive mobility by failure (RMF) mechanism. This mechanism means that when a certain specially declared Prolog predicate fails, the MoviLog machine transparently moves the Brainlet to another host which has declared the same predicate to try to satisfy the current goal. Tue, 01 Jan 2002 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/183165 2002-01-01T00:00:00Z Nowadays, Java-based platforms are the most common proposals for building mobile agent systems using web technology. However, the weak mobility model they use, the lack of adequate support for developing inference and reasoning mechanisms, added to the inherent complexity of building location aware software, impose strong limitations for developing mobile intelligent agent systems. In this article we present MoviLog, a platform for building Prolog-based mobile agents with a strong mobility model. MoviLog is an extension of JavaLog, an integration of Java and Prolog, that allows a user to take advantage of the best features of the programming paradigms they represent. MoviLog provides logic modules, called Brainlets, which are able to migrate among different web sites, either proactively or reactively, to use the available knowledge in order to find a solution. The most interesting feature introduced by MoviLog is the reactive mobility by failure (RMF) mechanism. This mechanism means that when a certain specially declared Prolog predicate fails, the MoviLog machine transparently moves the Brainlet to another host which has declared the same predicate to try to satisfy the current goal.