This paper resolves previous problems in the Multi-Strategy architecture for online learning of robotic behaviours. The hybrid method includes a symbolic qualitative planner that constructs an approximate solution to a control problem. The approximate solution provides constraints for a numerical optimisation algorithm, which is used to refine the qualitative plan into an operational policy. Introducing quantitative constraints into the planner gives previously unachievable domain independent reasoning. The method is demonstrated on a multi-tracked robot intended for urban search and rescue.

The paper introduces the ParkinsonCheck application. It is an app for smart phones based on spirography (spiral drawing) intended to detect signs of Parkinson's disease (PD) and essential tremor (ET), which is the main differential diagnosis from PD in the early stage of the disease. The app is equipped with an expert system and is the first such app to be completely automated. Its intended use is twofold: (a) to act as a standalone test for general population, advising potential patients to seek medical help as early as possible, and (b) to be used by neurologists as a portable and inexpensive fully digitalised clinical decision support system. ParkinsonCheck is currently freely available in Slovenia on four mobile platforms as a pilot study. After potentially upgrading its expert system with new learning data, the plan is for it to be translated into English and offered worldwide.

Qualitative Simulation (QSIM) reasons about the behaviour of dynamic physical systems as they evolve over time. The system is represented by a coarse qualitative model rather than precise numerical models. However, for large complex domains, such as robotics for Urban Search and Rescue, existing QSIM implementations are inefficient. ASPQSIM is a novel formulation of the QSIM algorithm in Answer Set Programming that takes advantage of the similarities between qualitative simulation and constraint satisfaction problems. ASPQSIM is compared against an existing QSIM implementation on a variety of domains that demonstrate ASPQSIM provides a significant improvement in efficiency especially on complex domains, and producing simulations in domains that are not solvable by the procedural implementation.

Inductive Logic Programming (ILP) is an area of Machine Learning which has now reached its twentieth year. Using the analogy of a human biography this paper recalls the development of the subject from its infancy through childhood and teenage years. We show how in each phase ILP has been characterised by an attempt to extend theory and implementations in tandem with the development of novel and challenging real-world applications. Lastly, by projection we suggest directions for research which will help the subject coming of age.

We present a novel approach to embodied learning of qualitative models. We introduce algorithm STRUDEL that enables an autonomous robot to discover new concepts by performing experiments in its environment. The robot collects data about its actions and its observations of the environment. From the obtained data, the robot learns qualitative descriptive models of the effects that its actions have in the environment. Models are learned using inductive logic programming. We describe two experiments with a humanoid robot Nao in which Nao learns descriptive qualitative models which contain what can be interpreted as simple definitions of the concepts of movability and stability.

We introduce a novel approach to cross-media learning based on argument based machine learning (ABML). ABML is a recent method that combines argumentation and machine learning from examples, and its main idea is to use arguments for some of the learning examples. Arguments are usually provided by a domain expert. In this paper, we present an alternative approach, where arguments used in ABML are automatically extracted from text with a technique for relation extraction. We demonstrate and evaluate the approach through a case study of learning to classify animals by using arguments automatically extracted from Wikipedia.

Knowledge elicitation is known to be a difficult task and thus a major bottleneck in building a knowledge base. Machine learning has long ago been proposed as a way to alleviate this problem. Machine learning usually helps the domain expert to uncover some of the more tacit concepts. However, the learned concepts are often hard to understand and hard to extend. A common view is that a combination of a domain expert and machine learning would yield the best results. Recently, argument based machine learning (ABML) has been introduced as a combination of argumentation and machine learning. Through argumentation, ABML enables the expert to articulate his knowledge easily and in a very natural way. ABML was shown to significantly improve the comprehensibility and accuracy of the learned concepts. This makes ABML a most natural tool for constructing a knowledge base. The present paper shows how this is accomplished through a case study of building a knowledge base of an expert system used in a chess tutoring application.

Recently we showed that under very reasonable conditions, incomplete, real-time search methods like RTA*work better with pessimistic heuristic functions than with optimistic, admissible heuristic functions of equal quality. The use of pessimistic heuristic functions results in higher percentage of correct decisions and in shorter solution lengths. We extend this result to learning RTA*(LRTA*) and demonstrate that the use of pessimistic instead of optimistic (or mixed) heuristic functions of equal quality results in much faster learning process at the cost of just marginally worse quality of converged solutions.

In this paper we discuss the application of a new machine learning approach - argumentation based machine learning - to the legal domain. Argumentation based machine learning is particularly suited to law as it makes use of the justifications of decisions to guide its learning. Importantly, where a large number of decided cases are available, it provides a way of identifying which need to be considered, so that only decisions which will have an influence are examined.

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.

UNLABELLED: Visual programming offers an intuitive means of combining known analysis and visualization methods into powerful applications. The system presented here enables users who are not programmers to manage microarray and genomic data flow and to customize their analyses by combining common data analysis tools to fit their needs. AVAILABILITY: http://www.ailab.si/supp/bi-visprog SUPPLEMENTARY INFORMATION: http://www.ailab.si/supp/bi-visprog.

UNLABELLED: VizRank is a tool that finds interesting two-dimensional projections of class-labeled data. When applied to multi-dimensional functional genomics datasets, VizRank can systematically find relevant biological patterns. AVAILABILITY: http://www.ailab.si/supp/bi-vizrank SUPPLEMENTARY INFORMATION: http://www.ailab.si/supp/bi-vizrank.

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 X2 distribution closely match those obtained with the boot-strap.

MOTIVATION: Genetic networks are often used in the analysis of biological phenomena. In classical genetics, they are constructed manually from experimental data on mutants. The field lacks formalism to guide such analysis, and accounting for all the data becomes complicated when large amounts of data are considered. RESULTS: We have developed GenePath, an intelligent assistant that automates the analysis of genetic data. GenePath employs expert-defined patterns to uncover gene relations from the data, and uses these relations as constraints in the search for a plausible genetic network. GenePath formalizes genetic data analysis, facilitates the consideration of all the available data in a consistent manner, and the examination of the large number of possible consequences of planned experiments. It also provides an explanation mechanism that traces every finding to the pertinent data. AVAILABILITY: GenePath can be accessed at http://genepath.org. SUPPLEMENTARY INFORMATION: Supplementary material is available at http://genepath.org/bi-.supp.

Controlling a complex dynamic system, such as a plane or a crane, usually requires a skilled operator. Such control skill is typically hard to reconstruct through introspection. Therefore an attractive approach to the reconstruction of control skill involves machine learning from operator's control traces, also known as behavioral cloning. In the most common approach to behavioral cloning, a controller is induced as a direct mapping from system states to actions. Unfortunately, such controllers usually suffer from lack of robustness and lack typical elements of human control strategies, such as subgoals and substages of the control plan. We investigate a novel approach. We apply the GoldHorn program to induce from the operator's trajectories a set of symbolic constraints. These are then used together with a locally weighted regression model to determine the next action. Using the Acrobot problem in a case study, this approach showed significant improvements both in terms of control performance and transparency of induced clones.

One way of enhancing the problem-solving power of a domain-specialised automatic problem solver is the introduction of domain-important concepts, defined as patterns, for streamlining the problem-solving process. The number of patterns needed for certain kinds of tasks, or domains, like chess, can be so high that the success of this approach to the development of powerful problem-solvers critically depends on two questions: <(a) How much programming effort is needed to implement such patterns in a programming language? (b) How efficiently, in terms of execution time, can these patterns be evaluated? In this paper several fundamentally different approaches to the implementation of problem-domain meaningful patterns, using a complex chess problem as an example, are compared. In particular, an approach based on cellular array processing operations is investigated.

This paper describes the first computer implementation of Master skill in a nontrivial chess end game other than by exhaustive tabulation.

I. Bratko, D. Kopec, D. Michie; Pattern-Based Representation of Chess End-Game Knowledge, The Computer Journal, Volume 21, Issue 2, 1 May 1978, Pages 149–1