Jure Žabkar

Argument based machine learning

We present a novel approach to machine learning, called ABML (argumentation based ML). This approach combines machine learning from examples with concepts from the field of argumentation. The idea is to provide experts arguments, or reasons, for some of the learning examples. We require that the theory induced from the examples explains the examples in terms of the given reasons. Thus arguments constrain the combinatorial search among possible hypotheses, and also direct the search towards hypotheses that are more comprehensible in the light of experts background knowledge. In this paper we realize the idea of ABML as rule learning. We implement ABCN2, an argument-based extension of the CN2 rule learning algorithm, conduct experiments and analyze its performance in comparison with the original CN2 algorithm.

Argument based machine learning applied to law

In this paper we discuss the application of a new machine learning approach – Argument Based Machine Learning – to the legal domain. An experiment using a dataset which has also been used in previous experiments with other learning techniques is described, and comparison with previous experiments made. We also tested this method for its robustness to noise in learning data. Argumentation based machine learning is particularly suited to the legal domain as it makes use of the justifications of decisions which are available. Importantly, where a large number of decided cases are available, it provides a way of identifying which need to be considered. Using this technique, only decisions which will have an influence on the rules being learned are examined.

Why is rule learning optimistic and how to correct it

In their search through a huge space of possible hypotheses, rule induction algorithms compare estimations of qualities of a large number of rules to find the one that appears to be best. This mechanism can easily find random patterns in the data which will – even though the estimating method itself may be unbiased (such as relative frequency) – have optimistically high quality estimates. It is generally believed that the problem, which eventually leads to overfitting, can be alleviated by using m-estimate of probability. We show that this can only partially mend the problem, and propose a novel solution to making the common rule evaluation functions account for multiple comparisons in the search. Experiments on artificial data sets and data sets from the UCI repository show a large improvement in accuracy of probability predictions and also a decent gain in AUC of the constructed models.

Automatic Spiral Analysis for Objective Assessment of Motor Symptoms in Parkinson's Disease

A challenge for the clinical management of advanced Parkinson’s disease (PD) patients is the emergence of fluctuations in motor performance, which represents a significant source of disability during activities of daily living of the patients. There is a lack of objective measurement of treatment effects for in-clinic and at-home use that can provide an overview of the treatment response. The objective of this paper was to develop a method for objective quantification of advanced PD motor symptoms related to off episodes and peak dose dyskinesia, using spiral data gathered by a touch screen telemetry device. More specifically, the aim was to objectively characterize motor symptoms (bradykinesia and dyskinesia), to help in automating the process of visual interpretation of movement anomalies in spirals as rated by movement disorder specialists. Digitized upper limb movement data of 65 advanced PD patients and 10 healthy (HE) subjects were recorded as they performed spiral drawing tasks on a touch screen device in their home environment settings. Several spatiotemporal features were extracted from the time series and used as inputs to machine learning methods. The methods were validated against ratings on animated spirals scored by four movement disorder specialists who visually assessed a set of kinematic features and the motor symptom. The ability of the method to discriminate between PD patients and HE subjects and the test-retest reliability of the computed scores were also evaluated. Computed scores correlated well with mean visual ratings of individual kinematic features. The best performing classifier (Multilayer Perceptron) classified the motor symptom (bradykinesia or dyskinesia) with an accuracy of 84% and area under the receiver operating characteristics curve of 0.86 in relation to visual classifications of the raters. In addition, the method provided high discriminating power when distinguishing between PD patients and HE subjects as well as had good test-retest reliability. This study demonstrated the potential of using digital spiral analysis for objective quantification of PD-specific and/or treatment-induced motor symptoms.

Argument-Based machine learning

In this paper, some recent ideas will be presented about making machine learning (ML) more effective through mechanisms of argumentation. In this sense, argument-based machine learning (ABML) is defined as a refinement of the usual definition of ML. In ABML, some learning examples are accompanied by arguments, that are experts reasons for believing why these examples are as they are. Thus ABML provides a natural way of introducing domain-specific prior knowledge in a way that is different from the traditional, general background knowledge. The task of ABML is to find a theory that explains the “argumented” examples by making reference to the given reasons. ABML, so defined, is motivated by the following advantages in comparison with standard learning from examples: (1) arguments impose constraints over the space of possible hypotheses, thus reducing search complexity, and (2) induced theories should make more sense to the expert. Ways of realising ABML by extending some existing ML techniques are discussed, and the aforementioned advantages of ABML are demonstrated experimentally.

Parkinson Check smart phone app

The paper introduces the Parkinson Check application. It is an app for smart phones based on spirography (spiral drawing) intended to detect signs of Parkinsons 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. Parkinson Check 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.

Application of argument based machine learning to law

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.

Extracting qualitative relations from categorical data

Qualitative modeling is traditionally concerned with the abstraction of numerical data. In numerical domains, partial derivatives describe the relation between the independent and dependent variable; qualitatively, they tell us the trend of the dependent variable. In this paper, we address the problem of extracting qualitative relations in categorical domains. We generalize the notion of partial derivative by defining the probabilistic discrete qualitative partial derivative (PDQ PD). PDQ PD is a qualitative relation between the target class c and the discrete attribute; the derivative corresponds to ordering the attributes values, a i , by P ( c | a i ) in a local neighborhood of the reference point, respecting the ceteris paribus principle. We present an algorithm for computation of PDQ PD from labeled attribute-based training data. Machine learning algorithms can then be used to induce models that explain the influence of the attributes values on the target class in different subspaces of the attribute space.

Qualitative Planning of Object Pushing by a Robot

Pushing is often used by robots as a simple way to manipulate the environment and has in the past been well studied from kinematic and numerical perspective. The paper proposes a qualitative approach to pushing convex polygonal objects by a simple wheeled robot through a single point contact. We show that by using qualitative reasoning, pushing dynamics can be described in concise and intuitive manner, that is still sufficient to control the robot to successfully manipulate objects. Using the QUIN program on numerical data collected by our robot while experimentally pushing objects of various shapes, we induce a model of pushing. This model is then used by our planning algorithm to push objects of previously unused shapes to given goal configurations. The produced trajectories are compared to smooth geometric solutions. Results show the correctness of our qualitative model of pushing and efficiency of the planning algorithm.

Extreme value correction: a method for correcting optimistic estimations in rule learning

Machine learning algorithms rely on their ability to evaluate the constructed hypotheses for choosing the optimal hypothesis during learning and assessing the quality of the model afterwards. Since these estimates, in particular the former ones, are based on the training data from which the hypotheses themselves were constructed, they are usually optimistic. The paper shows three different solutions; two for the artificial boundary cases with the smallest and the largest optimism and a general correction procedure called extreme value correction (EVC) based on extreme value distribution. We demonstrate the application of the technique to rule learning, specifically to estimating classification accuracy of a single rule, and evaluate it on an artificial data set and on a number of UCI data sets. We observed that the correction successfully improved the accuracy estimates. We also describe an approach for combining rules into a linear global classifier and show that using EVC estimates leads to more accurate classifiers.