Search

[EN] Monte Carlo Tree Search (MCTS) is a heuristic method used in decision-making that aims to simulate all possible outcomes that result from choosing a given action, assigning a value to each of them and choosing the best one. In this work, it was decided to apply it to the popular game UNO, a card game with hidden information and a great degree of uncertainty and randomness involved. MCTS had already been implemented for games with these characteristics, such as Poker, Bridge or Hearts, among others. We implement a simulator of the game for its two-player version. The MCTS agent and several simulation strategies and other tweaks are then integrated, and agents are matched against each other and against a virtual player with the objective of comparing the different versions and interpret the obtained results., [ES] El árbol de búsqueda Monte Carlo (MCTS) es un método heurístico usado en la toma de decisiones que trata de simular todos los posibles resultados que surgen de elegir una determinada acción, asignando un valor a cada una de ellas y eligiendo la mejor. En este trabajo, se ha decidido aplicar el método al popular juego UNO, un juego de cartas con información oculta y un alto grado de incertidumbre y aleatoriedad involucrados. MCTS ya había sido implementado para juegos de estas características, como el Poker, Bridge o Hearts, entre otros. Se ha implementado un simulador del juego para su versión de dos jugadores. El agente MCTS y varias estrategias de simulación han sido integradas en el mismo, y los agentes son enfrentados entre ellos y contra un jugador virtual con el objetivo de comparar las diferentes versiones e interpretar los resultados obtenidos., [CA] La recerca Monte Carlo (MCTS) es un mètode heurístic emprat en la presa de decisions que tracta de simular tots els possibles resultats que sorgixen de triar una determinada acció, assignant un valor a cada una d’elles i triant la millor. En aquest treball, s’ha decidit aplicar el mètode al popular joc UNO, un joc de cartes amb informació oculta i un alt grau d’incertesa i aleatorietat involucrats. MCTS ja havia sigut implementat per a jocs d’aquestes característiques, com ara el Poker, Bridge o Hearts, entre altres. S’ha implementat un simulador del joc per a la versió de dos jugadors. L’agent MCTS i diverses estratègies de simulació hi han sigut integrades, i els agents són enfrontats entre ells i contra un jugador virtual amb l’objectiu de comparar les diferents versions i interpretar els resultats obtinguts.

[EN] Genetic algorithms (GA) are a good technique for obtaining near-optimal solutions to problems with no known strategy, saving computational power. However, without the adequate configuration of parameters, a GA will not operate as efficiently as it should. An incorrect parameterization can avoid the algorithm to find good solutions. In this project we have designed, implemented and evaluated a GA and studied its parameterization for solving a combinatorial problem, specifically the Travelling Salesman Problem. We will, given a problem configuration, present the best GA parameterization; the one that obtains solutions with better quality while consuming less computational power., [ES] Los algoritmos genéticos (AG) son una buena técnica para obtener soluciones casi óptimas para problemas sin una estrategia conocida, ahorrando potencia de cálculo. Sin embargo, sin la configuración adecuada de los parámetros, un AG no funcionará tan eficientemente como debería. Una parametrización incorrecta puede evitar que el algoritmo encuentre buenas soluciones. En este proyecto hemos diseñado, implementado y evaluado un AG y estudiado su parametrización para resolver un problema combinatorio, específicamente el Problema del Viajante de Comercio. Dada una configuración de problema, presentaremos la mejor parametrización del AG; es decir, la que obtiene soluciones de mejor calidad y consume menos potencia de cálculo., [CA] Els algoritmes genètics (AG) són una bona tècnica per a obtindre solucions quasi òptimes per a problemes sense una estratègia coneguda, estalviant potència de càlcul. No obstant això, sense la configuració adequada dels paràmetres, un AG no funcionarà tan eficientment com deuria. Una parametrización incorrecta pot evitar que l’algoritme trobe bones solucions. En aquest projecte hem disenyat, implementat i evaluat un AG i estudiat la seua parametrización per a resoldre un problema combinatori, específicament el Problema del Viatjant de Comerç. Donada una configuració de problema, presentarem la millor parametrización de l’AG; és a dir, la que obté solucions de millor qualitat i consumix menys potència de càlcul.

[EN] Monte Carlo Tree Search (MCTS) is a heuristic method used in decision-making that aims to simulate all possible outcomes that result from choosing a given action, assigning a value to each of them and choosing the best one. In this work, it was decided to apply it to the popular game UNO, a card game with hidden information and a great degree of uncertainty and randomness involved. MCTS had already been implemented for games with these characteristics, such as Poker, Bridge or Hearts, among others. We implement a simulator of the game for its two-player version. The MCTS agent and several simulation strategies and other tweaks are then integrated, and agents are matched against each other and against a virtual player with the objective of comparing the different versions and interpret the obtained results., [ES] El árbol de búsqueda Monte Carlo (MCTS) es un método heurístico usado en la toma de decisiones que trata de simular todos los posibles resultados que surgen de elegir una determinada acción, asignando un valor a cada una de ellas y eligiendo la mejor. En este trabajo, se ha decidido aplicar el método al popular juego UNO, un juego de cartas con información oculta y un alto grado de incertidumbre y aleatoriedad involucrados. MCTS ya había sido implementado para juegos de estas características, como el Poker, Bridge o Hearts, entre otros. Se ha implementado un simulador del juego para su versión de dos jugadores. El agente MCTS y varias estrategias de simulación han sido integradas en el mismo, y los agentes son enfrentados entre ellos y contra un jugador virtual con el objetivo de comparar las diferentes versiones e interpretar los resultados obtenidos., [CA] La recerca Monte Carlo (MCTS) es un mètode heurístic emprat en la presa de decisions que tracta de simular tots els possibles resultats que sorgixen de triar una determinada acció, assignant un valor a cada una d’elles i triant la millor. En aquest treball, s’ha decidit aplicar el mètode al popular joc UNO, un joc de cartes amb informació oculta i un alt grau d’incertesa i aleatorietat involucrats. MCTS ja havia sigut implementat per a jocs d’aquestes característiques, com ara el Poker, Bridge o Hearts, entre altres. S’ha implementat un simulador del joc per a la versió de dos jugadors. L’agent MCTS i diverses estratègies de simulació hi han sigut integrades, i els agents són enfrontats entre ells i contra un jugador virtual amb l’objectiu de comparar les diferents versions i interpretar els resultats obtinguts.

[EN] Genetic algorithms (GA) are a good technique for obtaining near-optimal solutions to problems with no known strategy, saving computational power. However, without the adequate configuration of parameters, a GA will not operate as efficiently as it should. An incorrect parameterization can avoid the algorithm to find good solutions. In this project we have designed, implemented and evaluated a GA and studied its parameterization for solving a combinatorial problem, specifically the Travelling Salesman Problem. We will, given a problem configuration, present the best GA parameterization; the one that obtains solutions with better quality while consuming less computational power., [ES] Los algoritmos genéticos (AG) son una buena técnica para obtener soluciones casi óptimas para problemas sin una estrategia conocida, ahorrando potencia de cálculo. Sin embargo, sin la configuración adecuada de los parámetros, un AG no funcionará tan eficientemente como debería. Una parametrización incorrecta puede evitar que el algoritmo encuentre buenas soluciones. En este proyecto hemos diseñado, implementado y evaluado un AG y estudiado su parametrización para resolver un problema combinatorio, específicamente el Problema del Viajante de Comercio. Dada una configuración de problema, presentaremos la mejor parametrización del AG; es decir, la que obtiene soluciones de mejor calidad y consume menos potencia de cálculo., [CA] Els algoritmes genètics (AG) són una bona tècnica per a obtindre solucions quasi òptimes per a problemes sense una estratègia coneguda, estalviant potència de càlcul. No obstant això, sense la configuració adequada dels paràmetres, un AG no funcionarà tan eficientment com deuria. Una parametrización incorrecta pot evitar que l’algoritme trobe bones solucions. En aquest projecte hem disenyat, implementat i evaluat un AG i estudiat la seua parametrización per a resoldre un problema combinatori, específicament el Problema del Viatjant de Comerç. Donada una configuració de problema, presentarem la millor parametrización de l’AG; és a dir, la que obté solucions de millor qualitat i consumix menys potència de càlcul.

Objective: Parkinson's disease (PD) is currently incurable, however proper treatment can ease the symptoms and significantly improve the quality of life of patients. Since PD is a chronic disease, its efficient monitoring and management is very important. The objective of this paper was to investigate the feasibility of using the features and methodology of a spirography application, originally designed to detect early Parkinson's disease (PD) motoric symptoms, for automatically assessing motor symptoms of advanced PD patients experiencing motor fluctuations. More specifically, the aim was to objectively assess motor symptoms related to bradykinesias (slowness of movements occurring as a result of under-medication) and dyskinesias (involuntary movements occurring as a result of over-medication). Materials and methods: This work combined spirography data and clinical assessments from a longitudinal clinical study in Sweden with the features and pre-processing methodology of a Slovenian spirography application. The study involved 65 advanced PD patients and over 30,000 spiral-drawing measurements over the course of three years. Machine learning methods were used to learn to predict the "cause" (bradykinesia or dyskinesia) of upper limb motor dysfunctions as assessed by a clinician who observed animated spirals in a web interface. The classification model was also tested for comprehensibility. For this purpose a visualisation technique was used to present visual clues to clinicians as to which parts of the spiral drawing (or its animation) are important for the given classification. Results: Using the machine learning methods with feature descriptions and pre-processing from the Slovenian application resulted in 86% classification accuracy and over 0.90 AUC. The clinicians also rated the computer's visual explanations of its classifications as at least meaningful if not necessarily helpful in over 90% of the cases. Conclusions: The relatively high classification accura

We propose a simple yet potentially very effective way of visualizing trained support vector machines. Nomograms are an established model visualization technique that can graphically encode the complete model on a single page. The dimensionality of the visualization does not depend on the number of attributes, but merely on the properties of the kernel. To represent the effect of each predictive feature on the log odds ratio scale as required for the nomograms, we employ logistic regression to convert the distance from the separating hyperplane into a probability. Case studies on selected data sets show that for a technique thought to be a black-box, nomograms can clearly expose its internal structure. By providing an easy-to-interpret visualization the analysts can gain insight and study the effects of predictive factors.

In the paper we show that diagnostic classes in cancer gene expression data sets, which most often include thousands of features (genes), may be effectively separated with simple two-dimensional plots such as scatterplot and radviz graph. The principal innovation proposed in the paper is a method called VizRank, which is able to score and identify the best among possibly millions of candidate projections for visualizations. Compared to recently much applied techniques in the field of cancer genomics that include neural networks, support vector machines and various ensemble-based approaches, VizRank is fast and finds visualization models that can be easily examined and interpreted by domain experts. Our experiments on a number of gene expression data sets show that VizRank was always able to find data visualizations with a small number of (two to seven) genes and excellent class separation. In addition to providing grounds for gene expression cancer diagnosis, VizRank and its visualizations also identify small sets of relevant genes, uncover interesting gene interactions and point to outliers and potential misclassifications in cancer data sets.

Usual numerical learning methods are primarily concerned with finding a good numerical fit to data and often make predictions that do not correspond to qualitative laws in the domain of modelling or expert intuition. In contrast, the idea of $Q^2$ learning is to induce qualitative constraints from training data, and use the constraints to guide numerical regression. The resulting numerical predictions are consistent with a learned qualitative model which is beneficial in terms of explanation of phenomena in the modelled domain, and can also improve numerical accuracy. This paper proposes a method for combining the learning of qualitative constraints with an arbitrary numerical learner and explores the accuracy and explanation benefits of learning monotonic qualitative constraints in a number of domains. We show that $Q^2$ learning can correct for errors caused by the bias of the learning algorithm and discuss the potentials of similar hierarchical learning schemes.

Attribute interactions are the irreducible dependencies between attributes. Interactions underlie feature relevance and selection, the structure of joint probability and classification models: if and only if the attributes interact, they should be connected. While the issue of 2-way interactions, especially of those between an attribute and the label, has already been addressed, we introduce an operational definition of a generalized $n$-way interaction by highlighting two models: the reductionistic part-to-whole approximation, where the model of the whole is reconstructed from models of the parts, and the holistic reference model, where the whole is modelled directly. An interaction is deemed significant if these two models are significantly different. In this paper, we propose the Kirkwood superposition approximation for constructing part-to-whole approximations. To model data, we do not assume a particular structure of interactions, but instead construct the model by testing for the presence of interactions. The resulting map of significant interactions is a graphical model learned from the data. We confirm that the P-values computed with the assumption of the asymptotic chi^2 distribution closely match those obtained with the bootstrap.

We report on new experiments with machine learning in the reconstruction of human sub-cognitive skill. The particular problem considered is to generate a clone of a human pilot performing a flying task on a simulated aircraft. The work presented here uses the human behaviour to create constraints for a search process that results in a controller -- pilot's clone. Experiments in this paper indicate that this approach, called ``indirect controllers'', results in pilot clones that are, in comparison with those obtained with traditional ``direct controllers'', simpler, more robust and easier to understand. An important feature of indirect controllers in this paper is the use of qualitative constraints.

Interactions are patterns between several attributes in data that cannot be inferred from any subset of these attributes. While mutual information is a well-established approach to evaluating the interactions between two attributes, we surveyed its generalizations as to quantify interactions between several attributes. We have chosen McGill's interaction information, which has been independently rediscovered a number of times under various names in various disciplines, because of its many intuitively appealing properties. We apply interaction information to visually present the most important interactions of the data. Visualization of interactions has provided insight into the structure of data on a number of domains, identifying redundant attributes and opportunities for constructing new features, discovering unexpected regularities in data, and have helped during construction of predictive models; we illustrate the methods on numerous examples. A machine learning method that disregards interactions may get caught in two traps: myopia is caused by learning algorithms assuming independence in spite of interactions, whereas fragmentation arises from assuming an interaction in spite of independence., Comment: 30 pages, 11 figures. Changes from v2: improved bibliography

Many effective and efficient learning algorithms assume independence of attributes. They often perform well even in domains where this assumption is not really true. However, they may fail badly when the degree of attribute dependencies becomes critical. In this paper, we examine methods for detecting deviations from independence. These dependencies give rise to “interactions” between attributes which affect the performance of learning algorithms. We first formally define the degree of interaction between attributes through the deviation of the best possible “voting” classifier from the true relation between the class and the attributes in a domain. Then we propose a practical heuristic for detecting attribute interactions, called interaction gain. We experimentally investigate the suitability of interaction gain for handling attribute interactions in machine learning. We also propose visualization methods for graphical exploration of interactions in a domain.

The usual numerical learning methods, that are primarily concerned with finding a good numerical fit to the data, often make predictions that do not correspond to the qualitative mechanisms in the domain of modelling or a domain expert's intuition. Consistency of numerical predictions with a given qualitative model is helpful when a numerical model is used for explanation of phenomena in the modelled domain, but can also considerably improve numerical accuracy. In this paper we present a novel approach to numerical machine learning called Qfilter. Qfilter is a numerical regression method that can take into account qualitative background knowledge to give qualitatively faithful numerical prediction. The results on a set of domains including population dynamics show considerable prediction accuracy improvements compared to the usual numerical learners. As qualitative domain knowledge is often available in practice, Qfilter's ability to exploit such knowledge should be beneficial in many applications.

Many effective and efficient learning algorithms assume independence of attributes. They often perform well even in domains where this assumption is not really true. However, they may fail badly when the degree of attribute dependencies becomes critical. In this paper, we examine methods for detecting deviations from independence. These dependencies give rise to “interactions” between attributes which affect the performance of learning algorithms. We first formally define the degree of interaction between attributes through the deviation of the best possible “voting” classifier from the true relation between the class and the attributes in a domain. Then we propose a practical heuristic for detecting attribute interactions, called interaction gain. We experimentally investigate the suitability of interaction gain for handling attribute interactions in machine learning. We also propose visualization methods for graphical exploration of interactions in a domain.

This article presents the results of an empirical experiment designed to gain insight into what is the effect of the minimax algorithm on the evaluation function. The experiment’s simulations were performed upon the KRK chess endgame. Our results show that dependencies between evaluations of sibling nodes in a game tree and an abundance of possibilities to commit blunders present in the KRK endgame are not sufficient to explain the success of the minimax principle in practical game-playing as was previously believed. The article shows that minimax in combination with a noisy evaluation function introduces a bias into the backed-up evaluations and argues that this bias is what masked the effectiveness of the minimax in previous studies.

There is much empirical evidence about the success of naive Bayesian classification (NBC) in medical applications of attribute-based machine learning. NBC assumes conditional independence between attributes. In classification, such classifiers sum up the pieces of class-related evidence from individual attributes, independently of other attributes. The performance, however, deteriorates significantly when the “interactions” between attributes become critical. We propose an approach to handling attribute interactions within the framework of “voting” classifiers, such as NBC. We propose an operational test for detecting interactions in learning data and a procedure that takes the detected interactions into account while learning. This approach induces a structuring of the domain of attributes, it may lead to improved classifier’s performance and may provide useful novel information for the domain expert when interpreting the results of learning. We report on its application in data analysis and model construction for the prediction of clinical outcome in hip arthroplasty.

We present a web-based implementation of GenePath, an intelligent assistant tool for data analysis in functional genomics. GenePath considers mutant data and uses expert-defined patterns to find gene-to-gene or gene-to-outcome relations. It presents the results of analysis as genetic networks, wherein a set of genes has various influence on one another and on a biological outcome. In the paper, we particularly focus on its web-based interface and explanation mechanisms.

We present a method for learning qualitative constraint trees implemented in the program QUIN (qualitative induction). At difference with usual decision trees, the leaves of qualitative trees contain qualitative functional constraints as used in qualitative reasoning. We demonstrate the use of qualitative trees by their application to the difficult task of cloning the human skill of bicycle riding. The approach results in qualitative controllers that are straightforward to explain, and also enable optimization of corresponding quantitative controllers which carry out the task in the style similar to the human, but more effectively.

We present a method for learning qualitative constraint trees implemented in the program QUIN (qualitative induction). At difference with usual decision trees, the leaves of qualitative trees contain qualitative functional constraints as used in qualitative reasoning. We demonstrate the use of qualitative trees by their application to the difficult task of cloning the human skill of bicycle riding. The approach results in qualitative controllers that are straightforward to explain, and also enable optimization of corresponding quantitative controllers which carry out the task in the style similar to the human, but more effectively.

We present a method for learning qualitative constraint trees implemented in the program QUIN (qualitative induction). At difference with usual decision trees, the leaves of qualitative trees contain qualitative functional constraints as used in qualitative reasoning. We demonstrate the use of qualitative trees by their application to the difficult task of cloning the human skill of bicycle riding. The approach results in qualitative controllers that are straightforward to explain, and also enable optimization of corresponding quantitative controllers which carry out the task in the style similar to the human, but more effectively.

We present a method for learning qualitative constraint trees implemented in the program QUIN (qualitative induction). At difference with usual decision trees, the leaves of qualitative trees contain qualitative functional constraints as used in qualitative reasoning. We demonstrate the use of qualitative trees by their application to the difficult task of cloning the human skill of bicycle riding. The approach results in qualitative controllers that are straightforward to explain, and also enable optimization of corresponding quantitative controllers which carry out the task in the style similar to the human, but more effectively.

We present a method for learning qualitative constraint trees implemented in the program QUIN (qualitative induction). At difference with usual decision trees, the leaves of qualitative trees contain qualitative functional constraints as used in qualitative reasoning. We demonstrate the use of qualitative trees by their application to the difficult task of cloning the human skill of bicycle riding. The approach results in qualitative controllers that are straightforward to explain, and also enable optimization of corresponding quantitative controllers which carry out the task in the style similar to the human, but more effectively.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such a control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operators' control traces, also known as behavioural cloning. In the most common approach to behavioural cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually lack typical elements of human control strategies, such as subgoals or desired trajectory and do not replicate the robustness of the human control skill. In this paper we investigate a novel approach. We apply the GoldHorn program to induce from the control traces a set of symbolic constraints. Those constraints describe the operator's generalized trajectory and are then used together with locally weighted regression model to determine the next action. Using the crane problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones. Moreover, generalizing the trajectory into qualitative strategy shows the potentials of such an approach.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such a control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operators' control traces, also known as behavioural cloning. In the most common approach to behavioural cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually lack typical elements of human control strategies, such as subgoals or desired trajectory and do not replicate the robustness of the human control skill. In this paper we investigate a novel approach. We apply the GoldHorn program to induce from the control traces a set of symbolic constraints. Those constraints describe the operator's generalized trajectory and are then used together with locally weighted regression model to determine the next action. Using the crane problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones. Moreover, generalizing the trajectory into qualitative strategy shows the potentials of such an approach.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such a control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operators' control traces, also known as behavioural cloning. In the most common approach to behavioural cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually lack typical elements of human control strategies, such as subgoals or desired trajectory and do not replicate the robustness of the human control skill. In this paper we investigate a novel approach. We apply the GoldHorn program to induce from the control traces a set of symbolic constraints. Those constraints describe the operator's generalized trajectory and are then used together with locally weighted regression model to determine the next action. Using the crane problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones. Moreover, generalizing the trajectory into qualitative strategy shows the potentials of such an approach.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such a control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operators' control traces, also known as behavioural cloning. In the most common approach to behavioural cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually lack typical elements of human control strategies, such as subgoals or desired trajectory and do not replicate the robustness of the human control skill. In this paper we investigate a novel approach. We apply the GoldHorn program to induce from the control traces a set of symbolic constraints. Those constraints describe the operator's generalized trajectory and are then used together with locally weighted regression model to determine the next action. Using the crane problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones. Moreover, generalizing the trajectory into qualitative strategy shows the potentials of such an approach.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such a control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operators' control traces, also known as behavioural cloning. In the most common approach to behavioural cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually lack typical elements of human control strategies, such as subgoals or desired trajectory and do not replicate the robustness of the human control skill. In this paper we investigate a novel approach. We apply the GoldHorn program to induce from the control traces a set of symbolic constraints. Those constraints describe the operator's generalized trajectory and are then used together with locally weighted regression model to determine the next action. Using the crane problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones. Moreover, generalizing the trajectory into qualitative strategy shows the potentials of such an approach.

In order to investigate the possibility of synthesizing new knowledge at a cognitive level more advanced than that which unaided human experts can reach, an ultra-complex chess endgame (King and two Bishops against King and Knight) was chosen as an experimental testbed. As a preparation for investigating this possibility, support was provided by the chess scholar and specialist in the endgame, A.J. Roycroft, and by a complete tabulation of factual knowledge of the domain completed by Kenneth Thompson. It turned out that no reasonable amount of practice, nor access to a database of 20p million chess facts, could enable Roycroft to master the endgame (thus the endgame is authentically 'ultra-complex'). The next step is to try and instill articulate mastery into a machine system by the 'rules from examples' method of computer induction. Keywords: Cognitive Science, Problem solving, Decision making, Computer tutor systems., Availability: Pub. in ICCA Jnl., v10 n1 p3-13 Mar 1986. No copies furnished by DTIC/NTIS.