doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Článek

10.1162/pres_a_00356

Katedra aplikované matematiky

Virtual reality has been with us for several decades already, but we are still trying to find the right ways to control it. There are a lot of controllers with various purposes and means of input, each with its advantages and disadvantages, but also with specific ways to be handled. Our hands were the primary means of input for human-computer interaction for a long time. However, now we can use movements of our eyes, our feet or even our whole body to control the virtual environment, interact with it, or move from one place to another. We can achieve this with various controllers and wearable interfaces, like eye tracking, haptic suits or treadmills. There are numerous devices that we can choose from for every category, but sometimes it can be hard to pick the one that suits our intentions best. This article summarises all types of user interface controllers for virtual reality, with their main pros and cons and their comparison.

Stať ve sborníku

Katedra aplikované matematiky

Příspěvek se zabývá problematikou radikalizačního procesu v kontextech terorismu a případů extrémního násilí. Představuje projekt bezpečnostního výzkumu podporovaný ministerstvem vnitra České republiky Detekce radikalizace v kontextu ochrany obyvatelstva a měkkých cílů před násilnými incidenty. Textu informuje o metodě detekce indikátorů radikalizace u potencionálně rizikových osob za pomocí souboru indikátorů a softwarové aplikace

Stať ve sborníku

10.1109/ICVR51878.2021.9483819

Katedra aplikované matematiky

Finding a simple and precise way to control the virtual environment is one of the goals of a lot of human-computer interaction research. One of the approaches is using a Leap Motion optical sensor, which provides hand and finger tracking without the need for any hand-held device. However, the Leap Motion system currently supports only one sensor at a time. To overcome this limitation, we proposed a set of algorithms to combine the data from multiple Leap Motion sensors to increase the precision and the usability of hand tracking. First, we suggested a way how to improve the calibration of the current hand pose alignment proposed by Leap Motion. Then, we proposed an approach to fuse the tracking data from multiple Leap Motion sensors to provide more precise interaction with the virtual world. For this, we implemented our very own algorithm for computing the confidence level of the tracking data that can be used to distinguish which Leap Motion sensor detects the tracked hands best. We implemented those algorithms into our MultiLeap library. We also created two demo scenes that we used to validate the correctness of our work - one for evaluation of the fusing algorithms and one for mimicking the interaction with control panels in a helicopter cockpit.

Stať ve sborníku

10.1109/AIVR52153.2021.00021

Katedra aplikované matematiky

We present a concept that provides hand tracking for virtual and extended reality, only with the use of optical sensors, without the need for the user to hold any physical controller. In this article, we propose five new algorithms further to extend our previous research and the possibilities of the hand tracking system whilst also making it more precise. The first algorithm deals with the need to calibrate the tracking system. Thanks to the new approach, we improved tracking precision by 37% over our previous solution. The second algorithm allows us to compute the precision of the hand tracking data when multiple sensors are used. The third algorithm further improves the computation of hand tracking data confidence by correctly handling the edge cases, for example, when the tracked hand is at the edge of the sensor's field of view. The fourth algorithm provides a new way to fuse the hand tracking data by using only the hand tracking data with the highest hand tracking data confidence. The fifth algorithm deals with the issue when the optical sensor misclassifies the hand chirality.

Článek

10.26717/BJSTR.2020.28.004676

Katedra aplikované matematiky

Each patient has his/her own individual characteristics. It is given by his/her main diagnosis and its detailed description (e.g., in oncology by TNM classification, grading, histological type etc.), comorbidities, results of genetic markers, biomarkers and many other factors. Standard patient care is usually based on the results of large clinical trials. This approach ignores the patient’s individuality. Using patient profiling based on tumor characteristics, genetic markers and biomarkers, it is possible to identify the drug with the highest clinical benefit for small patient groups where, according to the guidelines, treatment choice is no longer defined only by diagnosis and TNM classification. This procedure belongs to the area of personalized medicine. An important methodology is how to find similar patients to a newcomer and how to evaluate clinical benefits and risks of different types of therapies in this group of similar patients. The main focus is on overall survival, but time to progression and adverse reactions to treatment are also important. Based on extensive patient data available in electronic medical records (hospital information systems), similar cases to a newcomer can be identified and some statistics for these similar patients can be provided.

Stať ve sborníku

10.1007/978-3-030-50423-6_17

Katedra aplikované matematiky

Missing data is one of the most common preprocessing problems. In this paper, we experimentally research the use of generative and non-generative models for feature reconstruction. Variational Autoencoder with Arbitrary Conditioning (VAEAC) and Generative Adversarial Imputation Network (GAIN) were researched as representatives of generative models, while the denoising autoencoder (DAE) represented non-generative models. Performance of the models is compared to traditional methods k-nearest neighbors (k-NN) and Multiple Imputation by Chained Equations (MICE). Moreover, we introduce WGAIN as the Wasserstein modification of GAIN, which turns out to be the best imputation model when the degree of missingness is less than or equal to 30%. Experiments were performed on real-world and artificial datasets with continuous features where different percentages of features, varying from 10% to 50%, were missing. Evaluation of algorithms was done by measuring the accuracy of the classification model previously trained on the uncorrupted dataset. The results show that GAIN and especially WGAIN are the best imputers regardless of the conditions. In general, they outperform or are comparative to MICE, k-NN, DAE, and VAEAC.

Stať ve sborníku

Katedra aplikované matematiky

One task that is often discussed in a computer vision is the mapping of an image from one domain to a corresponding image in another domain known as image-to-image translation. Currently there are several approaches solving this task. In this paper, we present an enhancement of the UNIT framework that aids in removing its main drawbacks. More specifically, we introduce an additional adversarial discriminator on the latent representation used instead of VAE, which enforces the latent space distributions of both domains to be similar. On MNIST and USPS domain adaptation tasks, this approach greatly outperforms competing approaches.

Stať ve sborníku

10.1007/978-3-030-22744-9_16

Katedra aplikované matematiky

In real-world applications, we can encounter situations when a well-trained model has to be used to predict from a damaged dataset. The damage caused by missing or corrupted values can be either on the level of individual instances or on the level of entire features. Both situations have a negative impact on the usability of the model on such a dataset. This paper focuses on the scenario where entire features are missing which can be understood as a specific case of transfer learning. Our aim is to experimentally research the influence of various imputation methods on the performance of several classification models. The imputation impact is researched on a combination of traditional methods such as k-NN, linear regression, and MICE compared to modern imputation methods such as multi-layer perceptron (MLP) and gradient boosted trees (XGBT). For linear regression, MLP, and XGBT we also propose two approaches to using them for multiple features imputation. The experiments were performed on both real world and artificial datasets with continuous features where different numbers of features, varying from one feature to 50%, were missing. The results show that MICE and linear regression are generally good imputers regardless of the conditions. On the other hand, the performance of MLP and XGBT is strongly dataset dependent. Their performance is the best in some cases, but more often they perform worse than MICE or linear regression.

Stať ve sborníku

10.1007/978-3-319-94809-6_1

Katedra aplikované matematiky

In practice we often encounter classification tasks. In order to solve these tasks, we need a sufficient amount of quality data for the construction of an accurate classification model. However, in some cases, the collection of quality data poses a demanding challenge in terms of time and finances. For example in the medical area, we encounter lack of data about patients. Transfer learning introduces the idea that a possible solution can be combining data from different domains represented by different feature spaces relating to the same task. We can also transfer knowledge from a different but related task that has been learned already. This overview focuses on the current progress in the novel area of asymmetric heterogeneous transfer learning. We discuss approaches and methods for solving these types of transfer learning tasks. Furthermore, we mention the most used metrics and the possibility of using metric or similarity learning.

Stať ve sborníku

10.5220/0006396700170027

Katedra teoretické informatiky

One of the main prerequisites in most machine learning and data mining tasks is that all available data originates from the same domain. In practice, we often can’t meet this requirement due to poor quality, unavailable data or missing data attributes (new task, e.g. cold-start problem). A possible solution can be the combination of data from different domains represented by different feature spaces, which relate to the same task. We can also transfer the knowledge from a different but related task that has been learned already. Such a solution is called transfer learning and it is very helpful in cases where collecting data is expensive, difficult or impossible. This overview focuses on the current progress in the new and unique area of transfer learning - asymmetric heterogeneous transfer learning. This type of transfer learning considers the same task solved using data from different feature spaces. Through suitable mappings between these different feature spaces we can get more data for solving data mining tasks. We discuss approaches and methods for solving this type of transfer learning tasks. Furthermore, we mention the most used metrics and the possibility of using metric or similarity learning.

Článek

10.1007/s00357-015-9174-2

Katedra teoretické informatiky

We propose to use the Zipfian distribution as a kernel for the design of a nonparametric classifier in contrast to the Gaussian distribution used in most kernel methods. We show that the Zipfian distribution takes into account multifractal nature of data and gives a true picture of scaling properties inherent in data. We also show that this new look at data structure can lead to a simple classifier that can, for some tasks, outperform more complex systems.

Článek

10.1109/TCYB.2014.2305697

Katedra teoretické informatiky

Correlation dimension (CD), singularity exponents, also called scaling exponents, are widely used in multifractal chaotic series analysis. CD and other measures of effective dimensionality are used for characterization of data in applications. A direct use of CD to multidimensional data classification has not been hitherto presented. There are observations that the correlation integral is a distribution function of distances between all pairs of data points, and that by using polynomial expansion of distance with exponent equal to the CD this distribution is transformed into locally uniform. The classifier is based on consideration that the influence of neighbor points of some class on the probability that the query point belongs to this class is inversely proportional to its distance to the CD, power. New classification approach is based on summing up all these influences for each class. We prove that a resulting formula gives an estimate of probability of the class, not a measure of membership to a class only, to which the query point belongs. For this assertion to be valid, it is necessary that exponent of the polynomial transformation must be the CD. We also propose an averaging approach that speeds up computation of the CD especially for large data sets. It is demonstrated that the CD-based classifier can outperform more sophisticated classifiers.

Stať ve sborníku

10.1063/1.4825492

Katedra teoretické informatiky

The Random Forest is a well-known method and also a program for data clustering and classification. Unfortunately, the original Random Forest program is rather difficult to use. Here we describe a new version of this program originally written in Fortran 77. The modified program in Fortran 95 needs to be compiled only once and information for different tasks is passed with help of arguments. The program was tested with 24 data sets from UCI MLR and results are available on the net.

Článek

10.14311/NNW.2013.23.028

Katedra teoretické informatiky

The GMDH MIA algorithm uses linear regression for adaptation. We show that Gauss-Markov conditions are not met here and thus estimations of network parameters are biased. To eliminate this we propose to use cloning of neuron parameters in the GMDH network with genetic selection and cloning (GMC GMDH) that can outperform other powerful methods. It is demonstrated on tasks from the Machine Learning Repository.

Kapitola v knize

10.5772/50787

Katedra teoretické informatiky

Many methods of Data Mining use data sets for setting their parameters, particularly training and testing sets. Setting of parameters corresponds to the learning (training) of the methods. It is e.g. a case of artificial neural networks and other adaptive (iterative) methods. Some of these methods utilize so-called validation set as well. A question that can arise is how to correctly divide or other way preprocess a given data set to these sets, i.e. how select data samples from the original set and place them into the training and testing sets. The chapter focuses on an overview of existing methods that deal with methods of data selection and sampling. A general approach to the problem of data selection to training, testing and eventually validation sets is discussed. To be able to compare individual approaches, model evaluation techniques are discussed as well. Data splitting is one of used approaches to construct training, testing and possibly validation sets, but there are many other approaches to proper data preparation. Instance selection methods and class balancing methods belong to these approaches and are discussed in the chapter. An overview of the methods together with their features, utilization, positives and negatives is given as well. Selected main principles are illustrated in figures.