ZOZNAM ÚSPEŠNÝCH PROJEKTOV VO VÝZVE 1/25:
Training of neural networks for imbalanced data classification
AUTOR: Peter Drotar
The classification of imbalanced datasets poses a substantial challenge in many real-world applications, from medical diagnostics and finance to industrial fault detection. Deep neural networks exhibit remarkable modeling capacity, yet their performance often deteriorates when trained on skewed class distributions. In this project, we propose an innovative framework that combines high-performance computing infrastructure with tailored oversampling and network architecture strategies to robustly address imbalanced learning.
Our approach leverages an HPC cluster to train deep neural networks under various imbalance scenarios at scale. By utilizing distributed training and parallel hyperparameter search, we efficiently explore a wide range of network configurations and optimization settings, which would be prohibitively time-consuming on conventional hardware. Additionally, we employ advanced minority oversampling methods and ensemble-based training protocols to ensure that newly generated synthetic samples effectively aid the network in learning critical minority-class characteristics without introducing excessive noise.
Experimental evaluations on multiple benchmark datasets with diverse imbalance ratios demonstrate that our HPC-driven methodology consistently outperforms standard deep learning baselines and traditional oversampling techniques, both in predictive accuracy and stability. The synergy between large-scale computational resources and refined sampling strategies leads to deeper exploration of model space and enhanced generalization capabilities. This project provides a scalable, flexible solution for practitioners aiming to harness the power of deep neural networks in imbalanced data settings, with promising applications in critical sectors such as healthcare, cybersecurity, and industrial monitoring.
Radiance Fields for Scalable Sensor Fusion and Video Compression in V2X Communication for Connected and Autonomous Vehicles
AUTOR: Matúč Dopiriak
This project aims to leverage radiance fields (RFs) for enhancing sensor fusion and video compression in V2X communication for connected and autonomous vehicles (CAVs). The objectives of this project are twofold:
Video Compression for V2X Communication Using RFs – To develop and implement RF-based video compression techniques that optimize data transmission in V2X communication while preserving critical scene information for downstream tasks such as object detection and trajectory prediction.
Sensor Fusion for Navigation Using RFs – To design and employ RFs for improved sensor fusion, integrating multimodal data from LiDARs, cameras, and radars to enhance real-time perception and decision-making in CAV navigation, particularly in dynamic and occluded environments.
2D Materials Modeling by Artificial Intelligence
AUTOR: Ján Brndiar
Nowadays when all electronic devices are shrinking two dimensional materials become the most promising candidate to deliver this goal. History teaches us that experimental material science is very complicated without strong synergy between theoretical concepts. But also on the theoretical part there exist cavities which prevent their usage globally in material science. In recent years machine learning has become a common part of computation modeling but not all possibilities of its usage in this field are explored. Main reasons are computer cost or precision of approach of underlay method to produce adequate datasets (quality and size). In this project we want to apply machine learning to fixed-node Quantum Monte Carlo (FNQMC) calculations with machine learning force field approach in study of dynamics of defects vacancies in molybdenum disulfide. Machine learning approach are crucial for FNQMC because comparing to the cheap density functional approach, computational price for FNQMC precision is 1000:1
Multi-dimensional joint inversion of geophysical data in areas with complex topography
AUTOR: Ján Vozár
An integrated study of MT, gravimetric, magnetic, and petrological data will be carried out to characterize the subsurface distribution of geophysical properties such as conductivity, density, and susceptibility, and their correlations within the models. The knowledge of geological structures and mainly tectonic dislocations (faults) affects various areas of problems (seismic risks, slope deformations, radon emanation and others) and provides important information about tectonic development of studied area mainly Tatras Mts. A second focus area is the mineralization zones and the Rožňava fault, which are situated in southern Slovakia in the Gemericum unit and mineralization zones in Scandinavia from Era.Min2 DRex project. The multidisciplinary research of geophysical and structural parameters and have been carried out in frame of APVV and VEGA projects with focus on the Western Carpathians.
Among other geophysical methods, magnetotellurics (MT) provides conductivity distribution within the Earth and is very good results in terms of identification of faults, fracture structures or deeper suture zones. Conductive minerals can also cause similar effects in fractures. The modelled data are in the mountains areas, where topographic effects have to be included in the modelling. We will used 3D codes ModEM and FEMTIC to model these MT data. For the calculation of joint inversion, we use the jif3d package, which can invert available MT, gravity, and magnetic data individually and together in any combination in 3D. When applying the algorithms in this study, different coupling strategies are selected to achieve the best resolution of the model and the greatest possible reduction of the interpretation ambiguity of the final model for the given geological structures and selected geophysical methods.
The Application of Neural Radiance Fields in the Autonomous Industry
AUTOR: Miroslav Imrich
This project focuses on the utilization of Neural Radiance Fields (NeRF) in the autonomous industry. The primary objectives are: to leverage NeRF for high-fidelity 3D scene reconstruction in autonomous robotics and transportation; to explore its integration in smart manufacturing for enhanced perception and digital twin applications and overall individual applications of NeRF in the autonomous industry.
By integrating NeRF into the autonomous robotics, manufactories and transportation, this project aims to enhance perception, navigation, and scene understanding, paving the way for more efficient and intelligent autonomous systems.
Automatic cranial implant 3D generation
AUTOR: Marek Ružička
This research project aims to develop an advanced AI-based solution for the automatic generation of 3D cranial implants using Graph Neural Networks (GNNs). The primary focus is to automate and optimize the complex task of designing patient-specific implants for cranial reconstruction, reducing the dependency on manual interventions. The project will leverage the SkullBreak dataset containing 570 skull samples and employ a sequential architecture involving a point cloud diffusion network and a voxelization network.
Due to the computational intensity of training these networks and conducting hyperparameter optimization, supercomputer resources equipped with 8x A100 or H100 GPUs are necessary. The scalable, parallel processing capabilities of GPUs will significantly accelerate training, reducing the duration of model development from 69 days on a single GPU to just 7 days per training cycle on a high-performance computing (HPC) cluster.
The proposed solution utilizes GNNs for their ability to efficiently model complex geometric relationships and capture spatial structures within 3D skull data. This innovative approach will generate implants that seamlessly fit cranial defects, improving surgical outcomes and advancing personalized medicine. By harnessing the combined power of state-of-the-art AI models and HPC environments, the project aims to revolutionize cranial implant design and contribute novel insights to the field of medical AI applications.
Machine Learning prediction of docking scores for SARS-CoV-2 proteins
AUTOR: Marián Gall
The integration of machine learning (ML) into computational chemistry and drug design has become pivotal for predicting molecular properties. Molecular docking remains the best available method for evaluating compound libraries and identifying potential drug candidates, enabling the assessment of compound affinities towards various protein targets. However, despite its effectiveness, molecular docking remains computationally demanding, especially when applied to large-scale screening. The analysis of thousands to millions of compounds requires significant computational resources, making high-throughput drug discovery an intensive and time-consuming process. As the scale of screening increases, optimizing computational efficiency becomes a crucial challenge in the field.
Machine learning methods, particularly neural networks (NNs), have proven to be well-suited for this task, offering near-instant evaluation of large compound datasets. Recent advances in artificial intelligence and the availability of high-performance computing (HPC) resources have enabled the application of state-of-the-art ML models to streamline drug development processes.
In this project, we aim to apply three different ML techniques to predict molecular docking scores, focusing on the inhibitory potential of three SARS-CoV-2 proteins: main protease (Mpro), papain-like protease (PLpro), and the spike glycoprotein. By utilizing machine learning, we plan to significantly reduce the computational time required for docking score prediction, highlighting the efficiency of this approach in the virtual screening process. The proposed work will demonstrate the potential of ML-augmented drug discovery process, supported by the computational power of HPC cluster Devana.
Triplet Photosensitizers for Two-Photon Excited Bioimaging and Photodynamic Therapy
AUTOR: Peter Hrobárik
Systems exhibiting strong two-photon absorption (TPA) play a crucial role in advanced laser technologies, including 3D microfabrication and high-capacity optical data storage. High TPA cross sections combined with excellent luminescence and/or singlet oxygen production can be achieved by designing relatively small S,N-heteroaromatic building blocks with strategically positioned electron-donating and electron-withdrawing functional groups. Additionally, complexing these charge-transfer-active molecules with selected metal ions further enhances their nonlinear optical (NLO) response. The most effective derivatives and substitution patterns can be identified through computer-aided screening, employing quantum-chemical calculations.
This project aims to functionalize recently developed TPA-active dyes to obtain other essential properties, such as increased solubility in polar media, introduction of spin-orbit coupling allowing spin-forbidden singlet-triplet transition and production of highly reactive singlet dioxygen and/or prolongation of excited-state lifetime to the microsecond time domain (transition from fluorescence to phosphorescence). These attributes, in conjunction with TPA activity, will facilitate the use of the developed systems in practical applications, particularly in high-resolution „multichannel“ nonlinear imaging of biological structures and photodynamic therapy using deeply penetrating near-infrared laser radiation.
The computational research will leverage cutting-edge quantum-chemical methods to predict both linear and nonlinear optical properties, as well as singlet-triplet transition efficiencies. Time-dependent density functional theory (TD-DFT) with local hybrids and long-range corrected functionals will be employed, alongside more advanced coupled-cluster methods such as RI-CC2, RI-ADC(2), and RI-ADC(3). These approaches incorporate essential dynamic and static electron correlation effects, ensuring a more accurate description of charge-transfer processes.
AutoXAI: A framework for optimizing explainable methods
AUTOR: Ján Kopčan
This project introduces AutoXAI, a novel framework for optimizing explainable AI methods. The framework addresses a fundamental challenge in explainability: balancing the trade-off between explanation fidelity and human comprehensibility. Drawing inspiration from AutoML concepts, AutoXAI automatically selects and configures explainability methods based on specific task requirements, model architecture, and input data characteristics. The framework employs multi-objective optimization and establishes quantitative metrics for measuring both fidelity and comprehensibility. The project will validate the framework’s effectiveness through human-in-the-loop testing within a claim-matching task for fact-checking applications. Expected outcomes include a comprehensive evaluation methodology, technical documentation, and potential applications in combating disinformation through improved model interpretability.
Tailoring High-Entropy Perovskites for Next-Generation Thermoelectrics
AUTOR: Subrahmanyam Bandaru
Perovskite oxide materials have been extensively studied for thermoelectric applications due to their low toxicity, eco-friendliness, and high elemental abundance, offering advantages over tin and lead chalcogenides. However, their relatively poor thermoelectric performance is hindered by low electrical conductivity and high thermal conductivity. This limitation arises from the intrinsically low phonon mean free paths in these perovskites, making it difficult to achieve significant thermal conductivity suppression. This project aims to compute various perovskite oxide structures and investigate the impact of doping with multiple elements at the A- and B-sites. The thermoelectric properties of doped perovskite oxides with varying molar ratios will be analyzed and compared to identify the most suitable candidate for thermoelectric applications.The general objective of the present work is to perform the extensive, systematical studies of the electronic structure calculations in the framework of the Density Functional Theory (DFT) of HEPs materials to select the best candidates of desirable thermoelectric properties. The theoretical study to analyze the electronic structure will be made using the computer code VASP based on the DFT. The electronic transport properties will be analyzed using the code BoltzTraP which is based on the Boltzmann transport theory within the constant relaxation time approximation. The lattice thermal conductivity will be analyzed by ShengBTE and/or almaBTE. HEPs with different compositions of dopants will be synthesized by the combination of mechanical alloying spark plasma sintering in parallel to the simulations.
Theoretical study of cooperative phenomena in strongly correlated electron and spin systems
AUTOR: Ľubomíra Regeciová
The proposed project is devoted to the theoretical study of cooperative phenomena in strongly correlated electron and spin systems. The special attention will be paid to specifying the key mechanisms which lead to formation and stabilization of inhomogeneous charge and spin ordering, superconductivity, itinerant ferromagnetism, ferroelectric and magnetocaloric phenomenon due to the big application potential of these phenomena and their possible coexistence. The study will be performed on comprehensive model, which will take into account all relevant interactions in rare-earth and transition metal compounds, where besides the spin-independent Coulomb interaction in d and f band also the spin dependent (double exchange) interaction between both bands will be
included. For a solution of this model we plan to elaborate new numerical methods, which will be subsequently used in combination with standard methods of quantum statistical physics (DMRG and QMC) to study the above mentioned phenomena.
Multithreaded Simulation of Evolutionary Dynamics
AUTOR: Branislav Brutovský
Evolutionary algorithms represent a prominent modelling and problem-solving tool, which draws attention in two large research fields – study of biological evolution and multiparameter optimization. Despite different focus in the respective areas, their applications in both fields require exploration of huge spaces of candidate solutions. The structure of evolutionary algorithms makes them well-suited for parallel implementation, meaning that one can significantly benefit from applying them on parallel computation infrastructure. Within the project, simulated evolutionary dynamics of the population of replicators (abstract “cells”) in variable environment will be implemented in multithreaded fashion on the HPC infrastructure.
Theoretical NMR spectroscopy – from NMR spectra to accurate magnetic moments
AUTOR: Andrej Hurajt
This project focuses on refining nuclear magnetic dipole moments through high-precision NMR shielding calculations. We aim to address the „hyperfine puzzle“ by rederiving the magnetic moments of 203Tl and 205Tl isotopes, and improving the magnetic dipole moments of 123Te, 125Te, and 77Se isotopes using modern NMR standards and ab initio methods. Additionally, we will apply machine-learning force fields to model NMR spectra for Na+, K+, and Mg2+ ions in ionic liquids, with comparisons to experimental data from CERN-ISOLDE and TRIUMF.
The crystal structure prediction of siloxane-based polymers using ab initio methods
AUTOR: Andrej Antušek
Crystal structure of polymers determines low temperature properties of polymers. We will develop a workflow for an automatic prediction of polymer crystal structures using DFT method. The workflow will be used for predicting crystal structures of novel siloxane-based polymers for special space applications.
Optimization methods for Quantum Computing
AUTOR: Martin Plesch
Quantum computing represents a paradigm shift in computational science, leveraging the principles of quantum mechanics — such as superposition, entanglement, and interference — to solve problems that are intractable for classical computers. Current quantum hardware, however, is limited by noise and scale in the so called Noisy Intermediate-Scale Quantum (NISQ) devices. The most promising algorithms for practical purposes seem to be hybrid once, as for instance the Variational Quantum Eigensolver (VQE), which calculates the smallest eigenvalue (ground energy) of a Hamiltonian.
Within the project, we shall further develop, test and tune classical optimization methods tailored for interplay with quantum computers in hybrid algorithms. In particular we will focus the HOPSO minimization method developed by our group specifically for implementation in VQE problems. The outputs of the project shall help to be fully prepared on the next generation of quantum computers that might be ready to provide practical advances in the field, even if compared to best classical computers.
Investigation of Knot Interactions on Supercoiled DNA (INVEKTIS)
AUTOR: Dušan Račko
The project focuses on leveraging High-Performance Computing to explore the biology and technology of DNA. Specifically, it investigates previously unexplored polymer entanglements and knot interactions in the presence of DNA supercoiling, using coarse-grained molecular models and simulations. These simulations are carried out with advanced molecular dynamics software such as ESPResSo, LAMMPS, and HOOMD, along with custom Monte Carlo algorithms. The project also employs cutting-edge topological analysis techniques using specialized software, including KymoKnot, Topoly, and Knoto-ID.
High-entropy perovskites materials for next-generation energy applications
AUTOR: Inga Zhukova
Perovskite materials have been used extensively in energy applications, including solid oxide cells, photovoltaics, batteries, and catalysis, demonstrating excellent performance. Perovskites have the general formula ABX3, where A is an alkali/alkaline earth metal or rare earth metal cation, B is a transition or a post-transition metal cation, and an oxygen or halogen anion is typically situated at the X-sites. Perovskite properties are often adjusted by substituting a few additional elements Into the A- or B-sites. Achieving concurrent high performance and stability is challenging. In fact, high performance requires the introduction of reactive transition metals, whereas strong stability demands less-reactive elements that enhance cation–oxygen bond strength. Thus, high-performance perovskites usually degrade quickly in adverse conditions, such as high temperature, high humidity, or strongly acidic or basic environments. To enhance perovskite stability, it is common to either substitute an inactive element into the perovskite or create a protective surface layer. However, both solutions often hamper electrochemical activity. Therefore, new perovskite design strategies to achieve high performance and stability are required for next-generation energy applications.
Deep Learning Distribution Framework for Autonomous Vehicles
AUTOR: Róbert Rauch
This research advances distributed computing for Connected and Autonomous Vehicles (CAVs) by optimizing the execution of resource-intensive deep learning models between vehicles and edge servers. We develop novel strategies for split computing, where model execution begins on the vehicle, pauses at an optimal intermediate layer, and completes on an edge server. Additionally, we implement early exits – intermediate branches that can terminate model execution when further processing becomes infeasible due to high latency.
Building on our previous work with CNNs, we first extend these distributed computing approaches to transformer architectures, moving from simple image classification to semantic segmentation. Our methodology employs LSTM networks and imitation learning to dynamically optimize the split points and exit decisions based on environmental conditions such as network speed and server load. Secondly, we explore the application of split computing and early exits to sensor fusion models, particularly focusing on LiDAR-based 3D segmentation – an approach that remains unexplored in current literature.
The research requires significant computational resources for training these sophisticated models and running extensive simulations. The results, targeted for publication in high-impact journals and well established conferences, will establish new benchmarks for distributed deep learning in CAV applications. We aim to demonstrate faster execution of complex computer vision tasks while maintaining acceptable model performance through tunable latency-performance trade-offs, ultimately enabling quicker vehicle response times.
Empirical and Computational Study of Organic Molecules in Diverse Solvents
AUTOR: Tomáš Ján Liška
The aim of this project, as outlined in testing project p1024-25-t, is to study the structure of substances in various solvents, which can differ significantly from their crystalline forms. Empirically determined structures, obtained through NMR spectroscopy, will be compared with those derived from ab initio computations. Computational work will be conducted using ORCA and Grimme lab software, employing robust tight-binding semi-empirical methods (GFN2-xTB), mGGA DFT methods (r2SCAN-3c), and (double) hybrid DFT methods (wB97X-D4, PBE0-D4, revDSD-PBEP86-D4). The empirical analysis provides essential structural insights, while ab initio computations will enable precise 3D structural determination by comparing computed and experimental chemical shifts.
Prediction of thermodynamically stable crystal structures of novel nickel oxides with the use of Evolutionary Algorithms and Density Functional Theory II.
AUTOR: Radovan Bujdák
Nickel oxides exhibit a wide range of applications in semiconductors, superconductors, energy storage, gas detection, and catalysis. While N iO is well-characterized,other reported phases such as Ni2O3 and N iO2 remain poorly understood, and certain possible stoichiometries (e.g. N i2O5) are completely absent from the Ni–O phase diagram. Recent discoveries of a zincblende-type structure in N iO suggest greater structural diversity than previously recognized. This study aims to predict and analyze novel Ni–O phases using Density Functional Theory (DFT) combined with Evolutionary Algorithms (EA). To date, we have generated and optimized thousands of possible structures, revealing significant information such as the formation of polyoxide anions in Ni2O5 and polytypism in N iO2. Our ongoing research focuses on further computational exploration of electronic, magnetic, and thermodynamic properties of selected structures, with the potential to uncover new insights into the Ni–O binary system and its applications.
Prediction of new binary palladium oxides using evolutionary algorithms and Density Functional Theory II.
AUTOR: Diana Fabušová
This research explores the structural diversity and properties of palladium oxides. While palladium and its monoxide (PdO) are crucial catalysts in automotive applications, particularly for CO oxidation and NOx reduction, other palladium oxide phases remain poorly understood. Although five palladium oxides (PdO, Pd2O3, PdO2, PdO3, and Pd2O) have been reported, only PdO has been extensively studied and commercially utilized. Our study employs a combination of Evolutionary algorithms and Density Functional Theory (DFT) to predict and characterize new binary palladium-oxygen phases, including Pd3O4, Pd4O5, and Pd5O4. Building upon our previous work on Pd2O, PdO3, and Pd2O5, which generated over 2 000 crystal structures for each stoichiometry, we are investigating their electronic properties and structural stability. Preliminary findings indicate the importance of spin polarization and electronic correlations for Pd2O on structural stability. This comprehensive investigation aims to expand our understanding of palladium oxides while potentially uncovering new technological potential.
Computational Modelling of Neurons: HCN and Calcium Channels in Excitability & Depression
AUTOR: Matúš Tomko
Depression and related neuropsychiatric disorders remain difficult to understand and treat, necessitating deeper insights into their biological mechanisms. This project investigates the hypothesis that increased excitability of hippocampal CA1 pyramidal neurons in offspring exposed to maternal stress results from altered expression of hyperpolarization-activated cyclic nucleotide-gated (HCN) and voltage-gated calcium channels.
Using biologically detailed computational modeling and a multi-parametric single-objective optimization technique, we will generate populations of morphologically and biophysically realistic conductance-based compartmental models for both stressed and control groups. These models will capture biophysical heterogeneities while aligning with electrophysiological constraints, allowing us to analyze the impact of maternal stress on ion conductances and key neuronal functions such as resonance, temporal coding, excitability, and energy efficiency.
To ensure robustness across structural variations, we will model 10 different neuronal morphologies, generating a total of 100 populations for both the control and stressed groups. This comprehensive dataset will provide valuable insights into how maternal stress affects neuronal behavior at the biophysical level.
Properties of advanced materials applicable in green technologies
AUTOR: Eva Scholtzová
The project’s primary scientific goal is to compare the adsorption properties of inorganic layered structures (ILS) based on the graphene, clays and their modifications affected by substitutions focusing on the study of the interactions and stability of pollutant-ILS complexes essential for eliminating pollutants from wastewater and soil. From an ecological point of view, knowledge of the keying mechanism of contaminants on the surface of the ILS is crucial. Structural stability, interaction energies, and mechanism of keying in the formation of complexes among the selected adsorbed pollutants will be studied for both ILS of the economically more affordable clay minerals and the more expensive materials based on graphene by computational methods (e.g., DFT-D3 method, ab initio molecular dynamics).
Grassland management identification based on object detection from orthoimagery
AUTOR: Marcel Hudcovič
Efficient management of grasslands is essential for maintaining biodiversity and providing ecosystem services because human activities influence the structure and function of grassland ecosystems. Management practices such as mowing or grazing are important for the sustainable use and conservation of the grasslands and preventing their transition towards different ecosystems. While field observations and surveys are effective at a small scale, these methods require labor and time, and their capacity to provide a complex view of grassland management across larger regions is limited. Very high spatial resolution imagery provides observation of detailed grassland structures and their changes. That will help in the more precise identification of different features concerning the management of grasslands. To identify detailed landscape structures, the high-resolution orthophotos represent suitable scales but their availability is limited in time for mapping of specific objects. However, the potential of AI-based object detection techniques enhanced the usefulness of orthophotos for this case by allowing the identification of specific objects such as livestock, machinery, or hay stacks, that directly link to the specific grassland management activities. By annotating these objects at orthophotos will enable us to accurately detect and classify management-related features with high precision and speed. This approach overcomes many of the practical limitations of manual identification and allows extensive mapping of grassland management practices with almost no human interventions. This study presents the outcomes of such an object-based detection model by integrating the YOLO algorithm applied over the country-wide orthophotomosaic of Slovakia. Within this study, we present the methodology of object detection over selected areas trained over manually outlined and further augmented training samples. The presented model is transferable and supposed to be applied elsewhere in similar landscapes wherever the high-resolution imagery is available. Obtained results extend the available information on grassland management strategies and may be used complemented with other data and remote sensing technologies to define the dimensions of grassland management regimes. Such knowledge is crucial for understanding the efficiency of biodiversity conservation efforts at multiple scales.
Design and numerical simulation of Josephson traveling-wave parametric amplifiers
AUTOR: Emil Rizvanov
Josephson junctions, as non-dissipative elements, facilitate achieving quantum-limited noise in Josephson parametric amplifiers. Their noise temperature can be an order of magnitude lower than that of HEMT amplifiers. Consequently, utilizing a Josephson amplifier with an amplification exceeding 10 dB as a preamplifier enables reducing the cryogenic HEMT amplifier’s noise temperature. In this study, we employ the novel program JoSIM and well-known WRspice to design a traveling wave parametric amplifier.
Patient-Specific In-Silico Modeling of Ventricular Electrical Activation Using the openCARP Platform
AUTOR: Lukáš Zelieska
This project focuses on patient-specific in-silico modeling of ventricular electrical activation during premature ventricular contractions (PVCs) using the openCARP platform. PVCs, a common arrhythmia, disrupt the heart’s rhythm and are often managed through catheter-based radiofrequency ablation (RFA). However, the success of RFA depends on the accurate identification of the ectopic focus, which is often challenging. To improve the localization of PVC origins and enhance diagnostic and therapeutic strategies, this project uses a reaction-diffusion bidomain model to simulate PVCs. By integrating electrocardiographic (ECG) data with patient-specific anatomical models, the simulations aim to predict PVC behavior and improve the accuracy of ablation procedures. The workflow includes data acquisition, segmentation of CT scans, 3D modeling, and simulation in the openCARP environment. Due to the computational complexity of these simulations, high-performance computing (HPC) resources are employed to accelerate the modeling process, enabling faster and more comprehensive simulations. The project’s goal is to enhance non-invasive diagnostic methods and personalized treatment planning by providing deeper insights into PVC mechanisms and their propagation, ultimately supporting clinicians in planning more effective therapeutic interventions.
Štúdium klasifikácie konečných grupoidov podľa počtu asociatívnych trojíc a ich aplikácie
AUTOR: Marián Kňazovický
V rôznych algebrických aplikáciach (konečné automaty, kryptológia,…) sa často stretávame s úlohou nájsť konečný grupoid, kde sú splnené určité vzťahy. Tak vzniká otázka, nakoľko je možné v tabuľke násobenia meniť prvky pri zachovaní asociatívnosti násobenia. Cieľom práce je vytvoriť prehľad o všetkých grupoidoch pre n=2,3,4 a pokúsiť sa vytvoriť predpoklady pre vytvorenie prehľadu pre n=5.
Electronic and optical properties of organic cations for OLED applications
AUTOR: Marián Matejdes
Developing organic optoelectronic materials with tailored photophysical properties is crucial for applications such as OLEDs and photovoltaic cells. Achieving optimal HOMO and LUMO energy levels remains challenging, particularly for stable blue emitters in OLEDs. This project employs DFT and TD-DFT to analyze the optical and electronic properties of xanthene, anthraquinone, indigo, cyanine, and phthalein derivatives, focusing on non-covalent interactions that lead to dimerization and aggregate formation. By optimizing geometries and simulating spectra, the obtained results provide insights into these systems‘ behavior and guide the design of efficient photoactive molecules for OLED applications.
Pd2O Thin-Film Stability on a-Al2O3 Substrate
AUTOR: Eva Pospisilova
There are two main challenges which I would like to address during the 2nd year of my post-doc at the Institute of Materials and Machine Mechanics (IMMM), SAS, Bratislava.
The first task concerns stability of Pd2O and Pt2O thin films of 10A thickness, which were allegedly experimentally deposited on a-Al2O3 substrate at 350-450C. The main aim is to confirm singular experimental observation of this ultrathin layer in the distant past [J. Kumar, R. Saxena, Journal of Less-Common Metals, 147, 1989, 59-71].
We would like to relax different sections and the respective terminations (such as O-terminated, Pd-terminated and PdO-terminated (001)-, (110)- and (111)-oriented sections) of the bulk antiferromagnetic Pd2O (and Pt2O) cubic crystal, and perform symmetry-breaking and Molecular Dynamics (MD) complemented with Machine-Learning Force Fields (MLFF) at various temperatures to assess their stability. We will also search for the associated magnetic states of these atomically-thin slabs, initializing our spin-polarized relaxations from various spin states.
We are to prepare representative a-Al2O3 bulk via MLFF annealings according to recipes described in the current DFT literature to cut it in order to produce proper substrate for our Pd2O thin films. Subsequently, we intend to study bonds at the a-Al2O3/Pd2O interface and the temperature stability thereof.
As Pd2O constitutes Pd-rich — and hence O-poor — relative of the well-known stable bulk phase PdO, next think we should investigate is the possibility of the spontaneous Pd2O formation on a-Al2O3 and the opportunity for its transformation in O atmosphere.
There are many other phenomena we might explore, since Pd2O layers are used as catalysts of chemical reactions. For instance, we may study adsorption of O2, N2, NO, CO, CH4, CH3, ethanol on the various stable Pd2O surfaces.
The second task deals with the metal (M) compacting by metal-oxide (MO) intergrain deposition. Experimentally, there exists technique called Atomic Layer Deposition [HITEMAL alloy: M. Balog et al., Materials Science & Engineering A, 613, 2014, 82-90], allowing to introduce thin layer MO of predefined thickness between metallic grains, that glues the M grains together. This way it was demonstrated that a-Al2O3 significantly improves Al tensile strength (enhances the respective modulus), temperature stability and its ductility/malleability/forgeability/plasticity as well, since a-Al2O3 perfectly adapts to Al grain boundaries and fracture always forms transgranularly.
What is more, it has been observed, that equivalent improvement of mechanical properties can be achieved by simple crystalline oxides, such as Zn/ZnO interface. Surprisingly, crystalline ZnO follows Zn grain boundaries as closely as a-Al2O3 those of fcc-Al. Consequently, a fracture of Zn/ZnO interface always happens transgranularly.
Extra, there are indications that Ni and Cu would have the same strengthening effect. On the contrary, TiO2 and MgO not do it, even when combined with Al matrix: the crack invariably happens along the M/MO interface, so the MO embrittles the metal. As for the combination Ti+a-Al2O3, Ti oxidates every time so the crack propagates regularly along Ti/TiO2 interface, regardless of the presence of a-Al2O3.
Though it is facile nowadays to study the mechanical properties experimentally (e.g. at IMMM), the respective microscopic mechanism of the strengthening effect behind it is not understood. By means of the standard DFT PAW PBE VASP setup we would like to visualize the change of the bonds at the M/MO interface subjected to perpendicular tensile stress at different temperatures and to compare various material implementations.
We wonder whether DFT is able to reproduce experimental differences between combinations of metals M and metal oxides MO at the interface and correctly predict the strengthening/embrittlement effect and uncover reasons behind it. If so, DFT should enable to forecast brand-new combinations of M/MO materials, blending even different metals on both sides of the interface.