National Repository of Grey Literature 684 records found  1 - 10nextend  jump to record: Search took 0.00 seconds. 
Predicting local structural properties from antibody sequence
Beňo, Roman ; Příhoda, David (advisor) ; Hoksza, David (referee)
Predicting local structural properties of antibodies at residual level is vital for detecting the presence of post-translational modifications (PTMs), which often induce structural change in the antibody, negatively impact its shelf-life and possibly lead to the loss of the therapeutic potential. In this work, we predict relative solvent accessibility (RSA) of individual residues. This property is, alongside with the type of amino acid in question, the key indicator for presence of methionine oxidations and other types of PTMs. Due to the conservation of the antibody structure, we identified that different classes of prediction methods yield almost interchangeable results - total mean absolute error (MAE) of 5.64 RSA percentage units measured for the best performing machine learning pipeline compared to the 5.96 measured for the best performing statistical pipeline. The significant prediction quality improvement observed within comparison to the random prediction method with MAE of 35.996 may be as well attributed to the sequence conservancy. In CDR regions, RSA values are harder to predict. Although the range of methods and procedures employed throughout this work is by far not able to yield complex structure predictions, it might constitute a modular, high-throughput tool to support one's choices when...
Porovnání open-source nástrojů pro strojové učení
Poliakova, Yevheniia
Poliakova, Y. Comparison of open-source tools for machine learning. Thesis. Brno: Mendel University in Brno, 2022. This work is devoted to the research of accessible open source artificial intelligence. The thesis describes a selected list of available artificial intelligence tools and the use of these tools for specific tasks. The main contribution of the work is the comparison of open-source tools using experiments focused on inductively controlled (supervised, classification) knowledge acquisition from large volumes of text and data. These experiments will be performed using selected open-source tools. The result of the work will be a conclusion about the advantages and disadvantages of the already mentioned platforms, their characteristics in solving specific problems and recommendations for choosing a platform according to the assigned task or data.
Automatická adaptace pouličního osvětlení na základě dat z kamer
Švanda, Jan
The thesis deals with the creation of a system that, based on the analysis of the real situation on the road using image data from cameras can provide data for the prediction of the optimal lighting settings for each moment. The thesis also deals with the selection and comparison of detection methods and subsequent optimization of the selected method.
Age prediction based on human face morphology
Žigová, Dominika ; Velemínská, Jana (advisor) ; Pilmann Kotěrová, Anežka (referee)
Age estimation is increasingly needed in numerous scientific disciplines, and thus the demand for appropriate age estimation methods is ever-growing. This master's thesis deals with age prediction based on the human facial morphology of people in the interval of 10 to 59 years. Three-dimensional virtual models of individuals of Czech, or Slovak, nationalities were used. The final sample for the thesis consists of 1046 3D facial scans, including 552 females and 494 males, for which age estimates were found using neural network models. Selected neural network models are based on two different approaches. While the PointNet, PointNet++, PointConv, and Xception networks use point clouds as input, the Multi- view Convolution Neural Network (MVCNN) utilizes multiple scan views. Point clouds were constructed from polygon meshes using uniform sampling of the mesh surface. In this case, models assess every single point. Therefore, a set containing the given object's 3D coordinates collected from its surface is obtained. Views of a particular scan result from recording a polygon mesh of the corresponding scan at a certain angle. This so-called multi-view approach is based on a projection, which records a 2D scan from various angles and then assesses and aggregates images into a general descriptor, which is...
Accuracy evaluation of neural network potentials for simulations of platinum nanocluster at hydroxylated silica interfaces
Pokorná, Kristýna ; Erlebach, Andreas (advisor) ; Vázquez Melis, Héctor (referee)
Supported platinum nanoparticles are important heterogeneous catalysts in many industrial processes, but their activity is strongly affected by particle diffusion and sintering mechanisms which lead to deactivation of the catalyst. To stabilise Pt nanoparticles, it is necessary to understand the reactive interactions of Pt with its support material, e.g., hydroxylated silica and defect-containing zeolites. Realistic simulations of such catalysts at the relevant timescales can be achieved with Neural Network Potentials (NNP) which retain ab initio accuracy at about 103 times lower computational costs compared to density functional theory (DFT) calculations. However, NNPs have only limited transferability to systems not included in the training database. Therefore, in this work recently developed SchNet NNPs were thoroughly tested. These NNPs were trained on a diverse set of Pt and defect-containing zeolites and hydroxylated silica surfaces. Firstly, the DFT database was extended by an active learning approach to accurately model the surfaces of α-quartz, MWW as well as the 2D zeolite layer IPC-1P (hydrolysis product of UTL). The NNPs trained on the new DFT database were then tested using MD simulations of systems unseen during the training procedure. These systems include a silanol nest containing...
Prediction of Czech GDP using mixed-frequency machine learning models
Kotlan, Ivan ; Polák, Petr (advisor) ; Kukačka, Jiří (referee)
The goal of this study is first to provide superior predictions of Czech GDP growth to the o cial estimates of the Czech Statistical O ce and the proxy estimation of the Czech National Bank. Secondly, to expand the literature that focuses on machine-learning predictions that utilizes data with various sampling frequency. Although in the first goal, this thesis did not succeed as all models, namely Ridge and Random Forest, failed to beat the predictions of o cial institutes, the thesis contributes to the yet scarce literature on mixed-frequency machine-learning prediction. Since no machine-learning model accounts for data with various frequencies, the thesis shows how to transform variables so that any machine-learning model can utilize them. Furthermore, di erent dataset modifications are explored, such as the prediction time: end of the reference quarter (nowcast) and 40 days after the reference quarter (backcast), standardized and non-standardized datasets. And finally, for the superior Ridge model, the e ect of so-called high-frequency variables (sampled every week) is explored. While Random Forest showed little e ect by using di erent versions of the dataset, in the case of the Ridge model, the type of dataset had a significant e ect. While the non-standardized Ridge produces better overall...
Supraglacial lakes detection and volume estimation from remote sensing data
Rusnák, Samo ; Brodský, Lukáš (advisor) ; Šobr, Miroslav (referee)
Supraglacial lakes detection and volume estimation from remote sensing data Abstract Supraglacial lakes play an important role in understanding glacier dynamics, including their response to climate change. This thesis explores the problematics of estimating lake depth and volume using a physical model. This brings challenges in considering the influence of various factors, such as cryoconite on glacier surface and suspended particular matter, which influences physical model, which is in research mostly neglected. Regression analysis of the g parameter of a physical model, representing light attenuation coefficient, and supervised classification of supraglacial lakes is applied in this thesis. The results reveal the variability of parameter Ad, representing lake bottom albedo reflectance, and its impact on predicted supraglacial lakes depth and volume. The results highlight the problem of global parameterisation of the physical model of supraglacial lakes and the need for further research to improve its accuracy and explore future possibilities in this field. Keywords: supraglacial lake, remote sensing, machine learning, physical model, depth estimation, regression analysis
Binning numerical variables in credit risk models
Mattanelli, Matyáš ; Baruník, Jozef (advisor) ; Teplý, Petr (referee)
This thesis investigates the effect of binning numerical variables on the per- formance of credit risk models. The differences are evaluated utilizing five publicly available data sets, six evaluation metrics, and a rigorous statistical test. The results suggest that the binning transformation has a positive and significant effect on the performance of logistic regression, feedforward artifi- cial neural network, and the Naïve Bayes classifier. The most affected aspect of model performance appears to be its ability to differentiate between eligible and ineligible customers. The obtained evidence is particularly pronounced for moderately-sized data sets. In addition, the findings are robust to the inclusion of missing values, the elimination of outliers, and the exclusion of categorical features. No significant positive effect of the binning transformation was found for the decision tree algorithm and the Random Forest model.
Pálková, Martina ; Hromada, Martin (referee) ; Řehák,, David (referee) ; Podroužek, Jan (advisor)
This thesis deals with the issue of machine learning in modeling of movement of pedestrians, the possibilities of its use and its limitations. The problem is shown on two real examples from practice. Digitization of industry and related use of advanced computing methods, such as artificial intelligence, has undergone unprecedented progress in the last ten years. Nevertheless, the construction industry lags behind other industries. This topic offers a lot of scope for research, the results of which can be very well applied in practice. Machine learning has the potential to be cheap and efficiently process large data sets with high accuracy, almost in real-time. In order for this to be possible, it is necessary to solve questions such as the appropriate choice for the given problem method, its architecture and parameter optimization. Another important direction of research is pre-processing of data, its format and division into training and test sets. This the topic, despite the great progress in the field of machine learning, is still discussed without uniform conclusions. These are the directions in which this work goes.
Automatic diagnosis of the 12-lead ECG using deep learning
Blaude, Ondřej ; Chmelík, Jiří (referee) ; Provazník, Valentine (advisor)
The aim of this diploma thesis is to investigate the problematics of automatic ECG diagnostics, namely on twelve-lead recordings. This problem is solved by standard methods such as random forest, artificial neural networks or K-nearest neighbors. However, thanks to its ability to independently extract symptoms, deep learning methods are also popular. All these methods are described in the theoretical part. In the practical part, deep learning models were designed, functionality support was verified using data from the PhysioNet database. Two pilot models were created and subsequently optimized. From the entire parameter optimization procedure, three models are available, of which the best accuracy achieves an F1 score of 87.35% and 83.7%, and the second best achieves an F1 score of 77.74% and an accuracy of 84.53%. The results achieved are discussed and compared with those of similar publications.

National Repository of Grey Literature : 684 records found   1 - 10nextend  jump to record:
Interested in being notified about new results for this query?
Subscribe to the RSS feed.