Petr Gajdoš

Comparison of Classification Algorithms for Physical Activity Recognition

The main aim of this work is to compare different algorithms for human physical activity recognition from accelerometric and gyroscopic data which are recorded by a smartphone. Three classification algorithms were compared: the Linear Discriminant Analysis, the Random Forest, and the K-Nearest Neighbours. For better classification performance, two feature extraction methods were tested: the Correlation Subset Evaluation Method and the Principal Component Analysis. The results of experiment were expressed by confusion matrixes.

Classification of cardiotocography records by random forest

The cardiotocography (CTG) is a diagnostic method which is widely used in prenatal care. The CTG is indicated since 27 weeks of pregnancy and it measures heart activity, uterine contraction and fetal movement. Results of the CTG allow recognizing of three basic different fetal states (physiological, suspect and pathological) and an obstetrician can determine a diagnosis and evaluate situation which can lead to the fetus death. The main aim of this work is to suggest and to test algorithm for automatic recognition of above mentioned states. This task is especially used in prenatal care as a support decision system.

Non-negative Matrix Factorization on GPU

Today, the need of large data collection processing increase. Such type of data can has very large dimension and hidden relationships. Analyzing this type of data leads to many errors and noise, therefore, dimension reduction techniques are applied. Many techniques of reduction were developed, e.g. SVD, SDD, PCA, ICA and NMF. Non-negative matrix factorization (NMF) has main advantage in processing of non-negative values which are easily interpretable as images, but other applications can be found in different areas as well. Both, data analysis and dimension reduction methods, need a lot of computation power. In these days, many algorithms are rewritten with the GPU utilization, because GPU brings massive parallel architecture and very good ratio between performance and price. This paper introduce computation of NMF on GPU using CUDA technology.

Correlation Network Analysis Reveals Relationships between MicroRNAs, Transcription Factor T-bet, and Deregulated Cytokine/Chemokine-Receptor Network in Pulmonary Sarcoidosis

Sarcoidosis is an inflammatory granulomatous disease with unknown etiology driven by cytokines and chemokines. There is limited information regarding the regulation of cytokine/chemokine-receptor network in bronchoalveolar lavage (BAL) cells in pulmonary sarcoidosis, suggesting contribution of miRNAs and transcription factors. We therefore investigated gene expression of 25 inflammation-related miRNAs, 27 cytokines/chemokines/receptors, and a Th1-transcription factor T-bet in unseparated BAL cells obtained from 48 sarcoidosis patients and 14 control subjects using quantitative RT-PCR. We then examined both miRNA-mRNA expressions to enrich relevant relationships. This first study on miRNAs in sarcoid BAL cells detected deregulation of miR-146a, miR-150, miR-202, miR-204, and miR-222 expression comparing to controls. Subanalysis revealed higher number of miR-155, let-7c transcripts in progressing (n = 20) comparing to regressing (n = 28) disease as assessed by 2-year follow-up. Correlation network analysis revealed relationships between microRNAs, transcription factor T-bet, and deregulated cytokine/chemokine-receptor network in sarcoid BAL cells. Furthermore, T-bet showed more pronounced regulatory capability to sarcoidosis-associated cytokines/chemokines/receptors than miRNAs, which may function rather as “fine-tuners” of cytokine/chemokine expression. Our correlation network study implies contribution of both microRNAs and Th1-transcription factor T-bet to the regulation of cytokine/chemokine-receptor network in BAL cells in sarcoidosis. Functional studies are needed to confirm biological relevance of the obtained relationships.

A new FCA algorithm enabling analyzing of complex and dynamic data sets

Analyzing data with the use of Formal Concept Analysis (FCA) enables complex insights into hidden relationships between objects and features in a studied system. Several improvements in this research area, such as Fuzzy FCA or L-Fuzzy Concepts, bring the possibility to analyze data with a certain rate of indeterminacy. However, the usage of FCA on larger complex data brings several problems relating to the time-complexities of FCA algorithms and the size of generated concept lattices. The fuzzyfication of FCA emphasizes the mentioned problems. This article describes significant improvements of a selected FCA algorithm. The primary focus was given on the system of an effective data storage. The binary data was stored with the use of finite automata that leads to the lower memory consumption. Moreover, the better querying performance was achieved. Next, we focused on the inner process of the computation of all formal concepts. All improvements were integrated into a new FCA algorithm that can be used to analyze more complex data sets.

Iris recognition on GPU with the usage of Non-Negative Matrix Factorization

In this paper, we describe an alternative method of the recognition of human irises with the usage of NonNegative Matrix Factorization. The proposed method has been implemented on graphic processor unit (GPU) which makes the method usable in the real world due to short computation time.

Normalization Impact on SVD-Based Iris Recognition

In this paper, we compare the performance of singular value decomposition for the recognition of human irises based for both unmodified irises and their normalized versions.

GPU Based Parallelism for Self-Organizing Map

Modern graphics cards take role of powerful computation hardware. This hardware becomes more popular due to purchasing costs and its availability. The advantages of Graphics Processor Unit (GPU) in parallel computation of Self-Organizing Network are described in this paper including a comparison with multi-threaded CPU. The parallelism on GPU is explained in a separated section. Mentioned section is divided into parts with respect to different forms of parallelism. The results of experiments at the end confirmed, that the utilization of GPU brings significant improvements in time of computation in case of large data sets.

Solving nearest neighbors problem on GPU to speed up the Fruchterman-Reingold graph layout algorithm

Fast searching of the nearest neigbors in unordered point clouds is a very common task. This article presents a new parallel method tested on graph layout algorithm. Graphs in computer science are widely used in social network analysis, computer networks, transportation networks, and many other areas. In general, they can visualize relationships between objects. However, fast drawing of graphs with readable layouts is still a challenge. This paper describes a novel variant of the Fruchterman-Reingold graph layout algorithm which is adapted to GPU parallel architecture using a new K-NN approach based on space-filling curves and a new way of repulsive forces computation on GPU. The paper contains both performance and quality tests of the algorithm.

Human activity recognition on raw sensor data via sparse approximation

Human physical activity monitoring is a relatively new problem drawing much attention over the last years due to its wide application in medicine, homecare systems, prisoner monitoring etc. This paper presents Orthogonal Matching Pursuit based classifier as a method for activity recognition and proposes a modification to the classifier that significantly increases recognition accuracy. Both methods show promising results in both total recognition and differentiation between certain activities achieving up to 99.60% recognition accuracy even without any prior data processing. A comparison with other methods is also provided.