The quantity of data generated increases daily, which makes it difficult to process. In the case of supervised learning, labeling training examples may represent an especially tedious and costly task. One of the aims of positive and unlabeled (PU) learning is to train a binary classifier from partially labeled data, representing a strategy for combining supervised and semi-supervised learning and alleviating the cost of labeling data fully. Still, the main strength of PU learning arises when the negative data are not directly available or too diverse. Although the generative approaches have shown promising results in this field, they also bring shortcomings, such as high computational cost, training instability, and inability to generate fully labeled datasets. In the paper, we propose a novel Conditional Generative PU framework (CGenPU) with a built-in auxiliary classifier. We develop a novel loss function to learn the distribution of positive and negative examples, which leads to a unique, desirable equilibrium under a nonparametric assumption. Our CGenPU is evaluated against existing generative approaches using both synthetic and real data. The characteristics of various methods, including ours, are depicted with different toy examples. The results demonstrate the state-of-the-art performance on standard positive and unlabeled learning benchmark datasets. Given only ten labeled CIFAR-10 examples, CGenPU achieves classification accuracy of 84%, while current state-of-the-art D-GAN framework achieves 54%. On top of that, CGenPU is the first single-stage generative framework for PU learning.

One of the biggest challenges in continual learning domains is the tendency of machine learning models to forget previously learned information over time. While overcoming this issue, the existing approaches often exploit large amounts of additional memory and apply model forgetting mitigation mechanisms which substantially prolong the training process. Therefore, we propose a novel SuperFormer method that alleviates model forgetting, while spending negligible additional memory and time. We tackle the continual learning challenges in a learning scenario, where we learn different tasks in a sequential order. We compare our method against several prominent continual learning methods, i.e., EWC, SI, MAS, GEM, PSP, etc. on a set of text classification tasks. We achieve the best average performance in terms of AUROC and AUPRC (0.7% and 0.9% gain on average, respectively) and the lowest training time among all the methods of comparison. On average, our method reduces the total training time by a factor of 5.4-8.5 in comparison to similarly performing methods. In terms of the additional memory, our method is on par with the most memory-efficient approaches.

We present an adaptation and application of frequent subgraph mining (FSM) in a time series of spatial multi-level directed graphs depicting probabilistic transitions of water masses between neighboring sea areas within a given time interval. The directed graphs are created from the results of the numerical model, the Mediterranean Ocean Forecasting System. We assign unique labels (geographical locations) to vertices of the multi-level directed graphs. Then, we add the edge labels as discretized values of the probabilities of transitions between vertices. This modification allows the use of the established algorithm gSpan to search for frequently directed subgraphs in the sequence of such directed graphs. Thus, we obtain both general and specific subgraphs, such as convergences, divergences, and paths of the ocean currents in the numerical model. The resulting substructures, revealed by directed subgraphs, match oceanographic structures (gyres, convergences/divergences, and paths) deduced from field observations, and can also serve as a tool for the validation of the numerical model of circulation in the sea.

This paper explores the idea of closed world specialisation (CWS). While traditional CWS is performed as a postprocessing step, we propose two different approaches to incorporating it into the induction process of a bottom-up inductive logic programming system. The motivation comes from the fact th at using CWS as a postprocessing step is incapable of solving problems in which the negated part of the hypothesis is crucial. We apply the proposed approaches to the ProGolem bottom-up ILP system. We give examples of problems, where classical CWS fails to find a complete and consistent solution, whereas the proposed approaches succeed. Tests on real-world datasets show that the proposed approaches perform at least as well as regular CWS, while being better in terms of predictive accuracy in some cases. We also point out some weaknesses of different CWS approaches.

In this article we compare and put to test two families of non-parametric approaches to constructing prediction intervals for arbitrary regression models in the supervised learning framework. It is often assumed for the errors to be independent and identically distributed, but we focus on the gener al case when the errors may be input dependent. The first family of approaches is based on the idea of explaining the total prediction error as a sum of the model's error and the error caused by noise inherent to the data, so the two are estimated independently. The second family is based on the assumption of similarity of the data and these approaches estimate the prediction intervals of the target regression variable by using sample's nearest neighbors. Results on a large set of artificial and real-world datasets show that one method from the second family is superior to other methods. Approaches from the first family always form valid, yet not necessarily confirmatory prediction intervals, whereas approaches from the second family prove to be more time efficient.

A probabilistic radial basis function (PRBF) network is an effective non-linear classifier. However, similar to most other neural network models it is non-transparent, which makes its predictions difficult to interpret. In this paper we show how a one-variable-at-a-time and an all-subsets explanation method can be modified for an equivalent and more efficient use with PRBF network classifiers. We use several artificial and real-life data sets to demonstrate the usefulness of the visualizations and explanations of the PRBF network classifier.

One of the most common causes of human death is stroke, which can be caused by carotid bifurcation stenosis. In our work, we aim at proposing a prototype of a medical expert system that could significantly aid medical experts to detect hemodynamic abnormalities (increased artery wall shear stress). Based on the acquired simulated data, we apply several methodologies for1) predicting magnitudes and locations of maximum wall shear stress in the artery, 2) estimating reliability of computed predictions, and 3) providing user-friendly explanation of the model's decision. The obtained results indicate that the evaluated methodologies can provide a useful tool for the given problem domain.

In Machine Learning, estimation of the predictive accuracy for a given model is most commonly approached by analyzing the average accuracy of the model. In general, the predictive models do not provide accuracy estimates for their individual predictions. The reliability estimates of individual pred ictions require the analysis of various model and instance properties. In the paper we make an overview of the approaches for estimation of individual prediction reliability. We start by summarizing three research fields, that provided ideas and motivation for our work: (a) approaches to perturbing learning data, (b) the usage of unlabeled data in supervised learning, and (c) the sensitivity analysis. The main part of the paper presents two classes of reliability estimation approaches and summarizes the relevant terminology, which is often used in this and related research fields.

We present a method for explaining predictions for individual instances. The presented approach is general and can be used with all classification models that output probabilities. It is based on decomposition of a model's predictions on individual contributions of each attribute. Our method works for so called black box models such as support vector machines, neural networks, and nearest neighbor algorithms as well as for ensemble methods, such as boosting and random forests. We demonstrate that the generated explanations closely follow the learned models and present a visualization technique which shows the utility of our approach and enables the comparison of different prediction methods.

The study verifies a hypothesis that GDV in fact measures vitality. For that purpose a limited study was performed and the findings support the hypothesis. However, further investigation is needed in order to more reliably confirm it.

The article deals with the problem of texture description. It presents a statistical approach. Specifically, it introduces the use of first- and second-order statistics on texture color spaces. At the end of the article, we also give estimations of the computational time complexities of the parameter calculations presented in this article and describe our experience on one application domain. This study is a preliminary preparation for the application of these methods to medical images.

This paper describes the prototype assembly planner, which derives the optimal assembly sequence from a product's specification. Assembly representation used as an input to the planner is based on spatial relationships between parts, which are directly related to the geometry of the assembly move. This way, in addition to the input specification of the product, they can be used for output specification of the necessary assembly moves. The space of all possible assembly sequences is represented by an AND OR graph. The graph is searched for the best sequence using the variant of the AO* heuristic algorithm. The output of the system is the optimal assembly sequence with a description of the required assembly moves. The assembly sequence is, at this stage, evaluated with respect to geometrical feasibility, stability, and parallelism. To cope with the complexity of the implemented approach, we developed efficient algorithms for generating and searching the problem space. Empirical evaluation shows the ability of the system to deal with real-world assemblies.

Ischaemic heart disease is one of the world's most important causes of mortality, so improvements and rationalization of diagnostic procedures would be very useful. The four diagnostic levels consist of evaluation of signs and symptoms of the disease and ECG (electrocardiogram) at rest, sequential ECG testing during the controlled exercise, myocardial scintigraphy and finally coronary angiography. The diagnostic process is stepwise and the results are interpreted hierarchically, i.e, the next step is necessary only if the results of the former are inconclusive. Because suggestibility is possible, the results of each step are interpreted individually and only the results of the highest step are valid. On the other hand, machine learning methods may be capable of objective interpretation of all available results for the same patient and in this way increase the diagnostic accuracy, sensitivity and specificity of each step. In the usual setting, the machine learning algorithms are tuned to maximize classification accuracy. In our case, the sensitivity and specificity were much more important, so we generalized the algorithms to take in account the variable misclassification costs. The costs can be tuned in order to bias the algorithms towards higher sensitivity or specificity. We conducted many experiments with four learning algorithms and different variations of our dataset (327 patients with completed diagnostic procedures). Our results show that improvements using machine learning techniques are reasonable and might find good use in practice.

Machine learning techniques can be used to extract knowledge from data stored in medical databases. In our application, various machine learning algorithms were used to extract diagnostic knowledge to support decisions in the diagnosis of sport injuries.