|
|
 |
Hybrid Event
|
| DS - Data Science |
S. Omanovic (University of Sarajevo, Sarajevo, Bosnia and Herzegovina), A. Midzic (University of Bihac, Bihac, Bosnia and Herzegovina), Z. Avdagic, D. Pozderac (University of Sarajevo, Sarajevo, Bosnia and Herzegovina), A. Toroman (University of Bihac, Bihac, Bosnia and Herzegovina)
Missing Values Interpolation in PurpleAir Sensor Data Based on a Correlation with Neighboring Locations Using KNIME Analytics Platform 
Missing values handling in any collected data is one of the first issues that must be resolved to be able to use that data. This paper presents an approach used for missing values interpolation in PurpleAir particle pollution sensor data, based on a correlation of the measurements from the observed locations with the measurements from its neighboring locations, using KNIME Analytics Platform. Results of our experiments with data from five locations in Bosnia & Herzegovina, presented in this paper, shows that this approach, which is relatively simple to implement, gives good results. All modeling and experiments were conducted using KNIME Analytics Platform.
|
M. Hoti, J. Ajdari, X. Zenuni, M. Hamiti (South East European University, Tetovo, Macedonia)
Spectral Analysis, Agglomerative, Mean Shift and Affinity Propagation Algorithms, Use on the Content from Social Media for Low-Resource Languages
Social networks, as part of our daily life, affect our behavior and lifestyle in various ways, making it important for each of us to be aware of their impact. Posts on social media platforms can have a profound effect on our mood, depending on our personal interpretation and opinion of them. Therefore, it is crucial to correctly classify these textual data to gain a better understanding of their impact. However, this task can be challenging, particularly when dealing with unlabeled data such as social media posts. An added challenge is working with low-resource languages. In this research, we investigate four unsupervised text clustering methods by testing them on a low-resource language, such as Albanian. The investigated algorithms are Spectral, Agglomerative, Mean Shift and Affinity Propagation, and by adjusting the working parameters, we tried to find a more appropriate application of them. Methods are applied to pre-processed data, textual posts, by use of different preprocessing techniques, and the results are presented and interpreted. This research aims to assist other researchers in the same field who have a specific focus on working with low-resource languages.
|
I. Kozjak, M. Mihelčić (Faculty of Science, Zagreb, Croatia)
Interactive Redescription Set Mining and Exploration
InterSet is a client-server web application that allows targeted and contextual redescription set exploration. The main drawback of this tool is that it uses fixed, precomputed sets of redescriptions to allow obtaining novel knowledge. In this work, we significantly extend the capabilities of InterSet by adding the possibility to create new redescriptions on predefined data. The main advantage of this is that the user can create new redescriptions in any step of exploration depending on the context and current research hypothesis, utilizing entity or attribute constraint-based redescription mining and enriching the existing set of redescriptions with newly obtained knowledge. This procedure further enhances the potential of exploring new research hypotheses while exploring large sets of redescriptions. The redescription mining addition to the InterSet tool is achieved by utilizing the CLUS-RM algorithm. Usefulness of the obtained environment for interactive redescription set creation, targeted and contextual exploration is demonstrated on real-world use case datasets.
|
E. Tata, J. Ajdari, N. Besimi (South East European University, Tetovo, Macedonia)
Fake News Detection: A Comprehensive Survey
The growth and increasing use of digital information platforms have dramatically changed the way news is produced, disseminated, and consumed in our society. Fake news can be found everywhere through popular platforms like social media and the internet. Efforts to develop an effective system for identifying fake news are numerous. Artificial intelligent tools are included to address this difficult issue. Fake news appears in different forms based on the features of their content. The aim of this research is to provide a comprehensive understanding of the various techniques within the domain of fake news detection through a systematic literature review of the existing work. This literature will demonstrate the most significant and relevant models to provide orientations in future research.
|
M. Urbanč, M. Depolli (Jožef Stefan Institute, Ljubljana, Slovenia)
Graphical User Interface to Perform Glacier Simulations with PISM
Computer simulations are used to study the dynamics of glaciers, to understand how glaciers respond to changes in climate, such as warming temperatures and increased precipitation. We present a product that resulted from an ongoing study of past glacial extent in the point where North Dinarides meet South-eastern Alps. This area is studied by a geological field work, complemented with computer simulations. Simulations are used to determine the local climate forcings that result in the creation of glaciers in the extent that is from geomorphological indices.
Simulations are performed with PISM (Parallel Ice Sheet Model), which is a very general and highly capable simulator for glacial dynamics. To tackle the complexities of ice dynamics, PISM employs several sub-models for various aspects and thus exposes a plethora of parameters. There are also several options for setting parameter values, i.e. through config files, command line options or input files. To get a good handle on all the provided buttons and knobs, we developed a user interface for setting parameters of out simulations and for executing PISM with the selected set of parameters on a remote workstation.
In this paper we present the GUI developed in Jupyter and deployed in a local JupyterLab installation. The environment contains DEM (Digital Elevation Map) files, Python scripts and other inputs locally but connects to a 128-core workstation for simulation execution. The simulation inputs and outputs are transferred between the computers over an SSH connection. The GUI is publicly exposed in a single-user environment, only for use, while development is performed on a git repository which is then synchronized to the public through git.
|
F. Strniša (Jožef Stefan Institute, Ljubljana, Slovenia), M. Vodopivec (National Institute of Biology, Piran, Slovenia), G. Kosec (Jožef Stefan Institute, Ljubljana, Slovenia)
Computational Performance Aspects of CROCO-BFM Coupling
Coastal and Regional Ocean COmmunity model (CROCO) is a modelling system used in oceanographic simulations. Its advanced algorithms for momentum, heat, and mass transport can be utilized via a relatively easy-to-use user interface. It, however, only comes with several basic biogeochemical models of its own, and the PISCES model for biogeochemistry, which it borrows from NEMO, a different oceanography modelling system. While PISCES offers some additional features over other pre-implemented biogeochemical models it is still rather basic in its essence. Biogeochemical Flux Model (BFM) is a dedicated biogeochemistry modelling system. Its complexity is user-defined, and customizable. It can be run as a standalone model or in conjunction with an oceanographic simulation. Previously we have reported on our work regarding the incorporation of BFM into CROCO. In this work we address the performance aspects of this coupling. Namely, compared to e.g. PISCES, BFM is a much more complex model with more tracers, which make it more computationally expensive. It is therefore of the essence to investigate, and to highlight possible areas of improvement in this coupling's computational performance.
Acknowledgments
The authors would like to acknowledge the financial support of Slovenian Research Agency (ARRS) in the framework of the research core funding No. P2-0095 and project J7-2599.
|
| Break |
| DS - Data Science |
E. Krishnasamy (University of Luxembourg, Belval, Luxembourg), I. Vasileska, L. Kos (University of Ljubljana, Ljubljana, Slovenia), P. Bouvry (University of Luxembourg, Belval, Luxembourg)
OpenMP Offloading and OpenACC Programming Module Approach for Object-oriented Plasma Device Algorithms
Presently, plasma physics is getting more important due to its applications in clean energy production (using fusion technology) and other fields such as chemical and material science. Even recently, Lawrence Livermore National Laboratory (LLNL) has demonstrated the capability of producing more energy through fusion compared to laser energy. So, there have to be more simulation experiments have to be conducted further to explore the advancement in plasma physics. One way of achieving this is that use the simulation using supercomputers. In this work, we parallelise a one-dimensional object-oriented plasma device algorithm, Object Oriented Plasma Device 1d (OOPD1), on a multicore CPU and GPU. We use the OpenMP programming model for the CPU version, and for the GPU, we use OpenMP offloading and OpenACC offloading. All of these approaches are compared to each other. Thus, it provides further suitable programming models with parallel capabilities for the existing OOPD1 to explore the available parallel architectures.
|
N. Mijić, D. Davidović (Ruđer Bošković Institute, Zagreb, Croatia)
Benchmark DPC++ Code and Performance Portability on Heterogeneous Architectures
Source code portability is becoming increasingly important in the development of new solutions in HPC due to the wide diversification of hardware and heterogeneity of systems. With Intel's oneAPI suite of programming tools and the Data Parallel C++ compiler, a single source code containing both host and device code can leverage hardware architectures from different vendors. Using compiler's interoperability, it can be linked to existing libraries such as MPI to run the program on a distributed memory system. In this paper, we analyze the performance of the distributed Cholesky-QR2 algorithm implemented both in CUDA C++ and Intel DPC++. The initial implementation shows promising results in terms of the portability of the code between different architectures such as CPUs and GPUs.
|
Z. Šojat, K. Skala (Ruđer Bošković Institute, Zagreb, Croatia)
A View into the Future of Academic Publishing
Throughout the history, as well as now, the development of human civilisation, as a collective, could only be based on the transfer of knowledge from those who thought, explored, collected, invented and tried to understand, towards those who will continue their work. The extreme amount of academic knowledge production in present times, compared to historical evolution, well shows the (natural) fact that the collective growth of knowledge is exponential. However, this extreme amount of academic work in all fields of human endeavour, and the consequent necessity to transfer the gained knowledge to the community through publishing, properly expressed through the Open Access movement, gave rise, in our times, to unexpected predatory behaviour on the side of many publishers. The publishing system suddenly came upside-down. Now authors have to pay the publishers for their work to be put on an open access web-page. And reviewers have to review the work to be published with no recompensation whatsoever, even regarding their reputation. So they work because of their sheer enthusiasm. Though some prospective authors get the money, needed to pay the publisher's web-page, from their institutions and projects (which is actually paid then from common taxes!), there is a huge intellectual potential in citizen scientists and, specifically, in retired academicians, who can not financially afford to transfer their knowledge, and continue their academic work, by publishing. (Only in the European Union there is approximately 2 million academicians who do not have the opportunity to do independent scientific work because they do not have a project-based or institutional basis for their work.) The other extremely negative side of the predatory publishing practices and the complete lack of transparency in the review and work ranking process, is the significant lowering of trustworthiness of published work, both from the aspect of the quality of work itself (if you pay for your work to be published, why would the publisher be interested in rejecting it?), as also from the aspect of improper categorisation in human fields (e.g. a symbolic description of a theory in the field of Religion Studies can not be accepted in a Mathematical journal!).
Therefore it is necessary to Democratise the process of Academic Publishing, by introducing modern information science based technologies, inter alias the Distributed Ledger and Distributed Database, as well as Virtual Money technologies. The DAP initiative and project aim to solve these present day problems by providing a free, democratic, trustworthy open access publishing platform with Virtual Money, the DAP Ergions, for recompensation of work, and recognition and ranking of high quality work and all DAP users. The development of DAP internal economy and the spread towards external economy using the Ergions, will provide a generic opportunity to all academicians, citizen scientists and the global population to be involved in the progress of gathering new human knowledge. Living during the human-caused 6th Mass Extinction, it is our responsibility as knowledge-gatherers and knowledge-growers to enable all human potential to be used towards the solution of those terrible Ecological problems we caused.
|
| RTA - Robotics Technologies and Applications |
T. Tadić, P. Ćurković (Faculty of Mechanical Engineering and Naval Architecture, Zagreb, Croatia)
Biped Robot Walking Based on Deep Reinforcement Learning |
J. Marić, L. Petrović, I. Marković (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Human Intention Recognition in Collaborative Environments Using RGB-D Camera |
M. Rossi (sees.ai, Chichester, United Kingdom), A. Jokić (Faculty of Mechanical Engineering and Naval Architecture, Zagreb, Croatia)
ROS Framework for Distributed Control of Networks of Dynamical Systems |
| BE - Biomedical Engineering |
I. Tomasic (Mälardalen University, Västerås, Sweden), A. Prkić (Private Dental Practice, Split, Croatia), A. Lesin, D. Kalibovic Govorko (School of Medicine, Split, Croatia), N. Tomasic (Karolinska University Hospital, Västerås, Sweden), I. Medvedec Mikic (School of Medicine, Split, Croatia)
Heart Rate Variability Monitoring with Savvy ECG Sensor during Dental Surgery
Surgical procedures are accompanied with anxiety and stress for both patients and surgeons. Currently pulse oximetry is used to monitor patients during dental procedures. This paper investigates the utility of various heart rate variability parameters, estimated from Savvy patch ECG recordings, obtained during surgical extraction of impacted lower third molars under local anesthesia, for the purpose of estimating psychological state of the patient, as well as pain levels, in real time. This could be particularly beneficial for chronic conditions that can deteriorate under psychophysical stress, such as cardiovascular diseases. Statistical analysis shows significant correlation between several heart rate variability parameters and changes in autonomous nervous system functionality. Additionally, there was a significant correlation between several variability parameters and patient reported anxiety and pain levels.
|
M. Gusev, N. Petrovski, L. Tonkovikj (Faculty of Computer Science and Engineering, Skopje, Macedonia)
Optimizing Heartbeat Classification Using Bezier Interpolation
Analog to digital conversion of electrocardiograms depends on the sampling frequency influencing the determination of a proper heartbeat location and precision of further digital processing. We set a research question to find the optimal number of interpolation points and reduce the mistakes in the similarity check of heartbeats and classification of ventricular beats. The experiments evaluate all neighbouring pairs of heartbeats from the standard benchmark MIT-BIH arrhythmia dataset resampled to a 125 Hz sampling frequency. The results show that even one more interpolation point, which corresponds to a sampling frequency of 250 Hz, will increase the performance versus the original 360 Hz sampling frequency. At the same time, the optimal is interpolation with additional five or seven points corresponding to 750 Hz, and 1000 Hz respectively. We found that a threshold value of 34 reveals the optimal performance to conclude a change between ventricular heartbeats and others, even in a 10-bit precision of the analog-digital conversion. The processing time and performance/cost-benefit analysis show that one interpolation point is the most beneficial.
|
D. Taralunga (Politehnica University of Bucharest, Bucharest, Romania)
Fetal Monitoring: Multi-channel Fetal ECG Denoising Based on Artificial Intelligence Approach
Continuous electronics fetal monitoring using cardiotocography (CTG) represents the standard of evaluating the health status of the fetus and the risk of the pregnancy, in developed countries. However, the CTG has many limitations: high false positive rates, can not be used for long term monitoring, poor sensitivity, it offers just the fetal heart rate and its variability etc. In this context, the fetal electrocardiogram (fECG) signal is used to obtain additional diagnostic information. On the other hand, the standard in clinical practice for obtaining the fECG is invasive, can pose a risk for both mother and fetus, can only be used during birth (very limited time window). An alternative is the abdominal fECG, that is recorded using a matrix of electrodes placed on the maternal abdomen. This approach is noninvasive and can be used for long term monitoring. The main drawback is the small signal to noise ratio for the abdominal fECG. Thus, the challenge is to isolate the fECG signal from other types of noise that are recorded by the abdominal electrodes: the maternal electrocardiogram (mECG), the electromyogram (EMG), the electrohysterogram (EHG), power line interference etc. In this paper the author proposes an algorithm based on artificial neural network approach to extract the fECG signal waveform from abdominal recorded signals. The database contains 4 channels of abdominal signals for each subject. A comparison is introduced, with other approaches described in literature for fECG denoising from abdominal signals.
|
M. Gusev, A. Shekerov (Faculty of Computer Science and Engineering, Skopje, Skopje, Macedonia), G. Temelkov (Innovation Dooel, Skopje, Macedonia), M. Jovanov (Faculty of Computer Science and Engineering, Skopje, Skopje, Macedonia)
ECG Compression Based on Successive Differences
In medicine, large amounts of data are gathered through various means, such as electronic medical records, clinical trials, and medical devices. One example of a medical device that generates a significant amount of data is the electrocardiogram (ECG) machine, which is used to measure the electrical activity of the heart. The ECG machine records the electrical signals produced by the heart and displays them as a graph, known as an ECG trace. This data can be used to diagnose and monitor various heart conditions, such as heart attacks, arrhythmias, and heart disease. The goal of this research is to reduce the amount of physical space the ECG data occupies, and to design an algorithm that not only reduces this space by a significant margin, but is fast, efficient and can run on hardware that is commonly found in a medical setting. The algorithm needs to be compatible with real-time transfer of ECG data across a network, Bluetooth, or other data transfer methods.
In this paper, we present an algorithm that compresses data gathered from an ECG. The algorithm helps with the transfer, storage, and real-time practical application of ECG data. With our algorithm, the storage space required to store the data can be reduced by more than half in some cases, and by 46% on average.
|
I. Kiprijanovska, F. Panchevski, S. Stankoski (Emteq Ltd., Skopje, Macedonia), M. Gjoreski (Faculty of Informatics, Lugano, Switzerland), J. Archer, J. Broulidakis, I. Mavridou (Emteq Ltd., Brighton, United Kingdom), B. Hayes (Christchurch Hospital, Christchurch, New Zealand), T. Guerreiro (Faculty of Sciences, Lisbon, Portugal), C. Nduka (Emteq Ltd., Brighton, United Kingdom), H. Gjoreski (Emteq Ltd., Skopje, Macedonia)
Smart Glasses for Gait Analysis of Parkinson’s Disease Patients
Parkinson’s disease (PD) is one of the most common neurodegenerative disorders of the central nervous system, which predominantly affects patients’ motor functions, movement, and stability. Monitoring movement in patients with PD is crucial for inferring motor state fluctuations throughout daily life activities, which aids in disease progression analysis and assessing patients' response to medications. In recent years, there has been an increase in the usage of wearable sensors for PD symptom monitoring. In this study, we present a preliminary analysis of smart glasses equipped with IMU sensors to provide objective information on the motor state in patients with PD. Data were collected from seven patients with varying levels of symptom severity. The patients performed the Timed-Up-and-Go (TUG) Test while wearing IMU-equipped glasses. Our analysis indicates that smart glasses can provide information about patients’ gait that can be used to assess the severity level of PD as measured by two standardized questionnaires. Furthermore, patient-specific clusters can be easily detected in the sensor data, hinting at the development of personalized models for patient-specific monitoring of symptom progression. Therefore, smart glasses have the potential to be used as an unobtrusive and continuous screening tool for PD patients’ gait, enhancing medical assessment and treatment.
|
| Break |
| BE - Biomedical Engineering |
M. Melinščak (Zagreb University of Applied Sciences, Zagreb, Croatia)
Attention-based U-net: Joint Segmentation of Layers and Fluids from Retinal OCT Images
Since its introduction in 2015, U-net has become state-of-the-art neural network architecture for biomedical image segmentation. Although many modifications have been proposed, few novel concepts were introduced. In recent years, some significant breakthroughs have been achieved by introducing attention or, more specifically, Transformers. Primarily Transformers were used for natural language processing (NLP), and mostly they gained popularity due to the latest large language models (LLM) applications. Many attempts to incorporate self-attention mechanisms into solving computer vision tasks resulted in Vision Transformer (ViT). As ViT has some downsides compared to convolutional networks (CNNs), neural networks which merge advantages from both concepts prevail, especially in small data regimes we often face in medicine. U-net architecture still outperforms ViT models as their high accuracy relies on vast amounts of data. This paper investigates how attention added in U-net architecture affects results. We evaluate the outcomes on a publicly available dataset which consists of 1136 retinal optical coherence tomography (OCT) images from 24 patients suffering from neovascular age-related macular degeneration (nAMD). Also, we compare results to previously published results, and it could be noted that attention-based U-net achieves higher Dice scores by a significant margin. The code is publicly available.
|
S. Stankoski (Emteq Ltd., Skopje, Macedonia), B. Sazdov, J. Broulidakis (Emteq Ltd., Brighton, United Kingdom), I. Kiprijanovska (Emteq Ltd., Skopje, Macedonia), B. Sofronievski, S. Cox (Emteq Ltd., Brighton, United Kingdom), M. Gjoreski (Faculty of Informatics, Lugano, Switzerland), J. Archer, C. Nduka (Emteq Ltd., Brighton, United Kingdom), H. Gjoreski (Emteq Ltd., Skopje, Macedonia)
Recognizing Activities of Daily Living Using Multi-sensor Smart Glasses
This paper describes the multi-sensor OCOSensetm smart glasses and their ability to recognize everyday activities. In particular, we used a machine-learning approach to recognize seven daily life activities using the 3-axis IMU (accelerometer, gyroscope, and magnetometer) and the barometer in the glasses. The data used in the experiments were collected from 24 participants who performed pre-defined activities while wearing the glasses. The method processes the stream of sensor data and recognizes the activity of the user wearing the glasses every two seconds. The best-performing model based on the XGBoost algorithm achieved an accuracy of 84.7% and an F1 score of 83.7%, with the most problematic categories being standing vs. sitting. In separate experiments, we analyzed transitional activities (sitting down, sitting up). The results were promising, and we plan to use these two activities as a temporal context to improve the recognition of standing vs. sitting. Finally, we analyzed the lying activity separately, i.e., to recognize on which side the user is lying. In these experiments, the model recognized the four laying orientations (up/down/left/right) with a 99.5% F1 score.
|
L. Vrček (Faculty of Electrical Engineering and Computing, Zagreb, Croatia), X. Bresson (National University of Singapore, Singapore, Singapore), T. Laurent (Loyola Marymount University, Los Angeles, California, United States), M. Schmitz (National University of Singapore, Singapore, Singapore), M. Šikić (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Reconstruction of Short Genomic Sequences with Graph Convolutional Networks
Genome reconstruction, without prior knowledge about the sequence we are reconstructing, is performed with tools called de novo genome assemblers. These tools rely on numerous heuristics and usually provide a fragmented reconstruction, even for sequences shorter than the entire genomes or chromosomes. One of the most common approaches to de novo assembly, called Overlap-Layout-Consensus (OLC), constructs a graph from short overlapping fragments, which heuristics then simplify and find a path through. In this work, we explore how graph neural networks (GNNs) can assist with this task, and show that the GNN-based Layout phase can reconstruct longer sequences than naive search algorithms or heuristics deployed in de novo assemblers, with no significant difference in compute time on sequences up to 10 Mbp in length.
|
B. Borozan, L. Borozan, D. Ševerdija, D. Matijević (Department of Mathematics, University of Osijek, Osijek, Croatia), S. Canzar (Department of Computer Science and Engineering, Pennsylvania State University, University Park, United States)
Fortuna Detects Novel Splicing in Drosophila scRNASeq Data
Recent developments in single cell RNA sequencing techniques (scRNASeq) have made large quantities of sequenced data available across numerous species and tissues. Alternative splicing (AS) of pre-mRNA introns varies between tissues and even between cell-types and can be altered in disease. The study of novel AS, using standard RNASeq data, has been extensively studied for many years, while similar work on scRNASeq data has been scarce, despite its potential to offer a broader insight into cell-type specific processes. In this paper, we propose a novel pipeline that uses fortuna, a method that efficiently classifies and quantifies novel AS events, to process scRNASeq samples. Due to its short lifespan, high number of progeny, low maintenance cost, and intricate alternative splicing patterns similar in complexity to those of mammals, Drosophila (fruit fly) is a species of particular interest to researchers. Therefore, we experimentally evaluate our pipeline on real-world Drosophila single cell data samples from the Fly Cell Atlas.
|
| Break |
| BE - Biomedical Engineering |
J. Lipovac, K. Križanović (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Using De Novo Metagenome Assembly for Improved Metagenomic Classification
Metagenomics is a rapidly growing field that allows for studying complex microbial communities. One of the first steps in the metagenomic analysis is the classification of the organisms present in a sample. This is usually done by comparing sequencing reads to a database of known organisms. With the recent development of long-read sequencing technologies, such as PacBio and Oxford Nanopore Technologies (ONT), it is now possible to generate highly accurate assemblies of genomes from metagenomic samples. This is typically done using a combination of reference-based and de novo assembly approaches. Assembling the genomes from the metagenomic sample, prior to classification, could improve classification results and also aid in identifying new, previously unknown species. However, the evaluation of metagenome assemblies is a challenging task and it is important to assess the quality of the assemblies in order to ensure the accuracy of downstream analyses. In this paper, we provide a detailed overview of metagenomic classification, de novo metagenome assembly process, and evaluation of metagenome assembly, highlighting various tools and techniques currently available for each step. We also present initial results showing that metagenomic classification can benefit from a previously assembled metagenome.
|
V. Vodilovska, S. Gievska, I. Ivanoska (Faculty of Computer Science and Engineering, Ss Cyril and Methodius University, Skopje, Macedonia)
Hyperparameter Optimization of Graph Neural Networks for mRNA Degradation Prediction
Graph Neural Networks (GNNs) emerged as increasingly attractive deep learning models for complex data, making them extremely useful in biochemical and pharmaceutical domains. However, building a good-performing GNN requires lots of parameter choices and Hyperparameter Optimization (HPO) can aid in exploring solutions.
This study presents a comparative analysis of several strategies for Hyperparameter Optimization of GNNs. The explored optimization techniques include complex algorithms such as the bio-inspired Genetic Algorithm, Particle Swarm Optimization, and Artificial Bee Colony. In addition, Hill Climb and Simulated Annealing as well as the commonly used methods Random Search and Bayesian Search have also been covered.
The proposed optimization algorithms have been evaluated on improving the performance of the GNN-based architectures developed for predicting mRNA degradation. The Stanford OpenVaccine dataset for mRNA degradation prediction has been used for training and testing the predictive models. Finding mRNA molecules with low degradation rates is an important issue when developing mRNA vaccines for diseases such as COVID-19.
The analysis shows promising HPO results for the Genetic Algorithm and Particle Swarm Optimization as well as the Random Search method with certain limitations, however, related to the data quality and the complexity of the evaluated algorithms.
|
G. Mirceva, A. Naumoski, A. Kulakov (Faculty of computer science and engineering, Skopje, Macedonia)
Examination of Different Representations of Proteins Using Protein Ray-based Descriptor and Deep Learning Models
The study of proteins has been of high importance because it is needed to understand the processes in the living organisms in which these molecules are involved. Proteomics is the research area that studies the protein structures. One of the tasks on which proteomics is focused on is solving the protein classification task. Although there are many studies focused on this problem, it is still a popular task because there is still need for faster methods for protein classification. The aim of the study presented in this paper is to develop a fast and accurate protein classification model. For that purpose, for feature extraction we use our protein ray-based descriptor. We use a deep learning architecture for generating prediction model. Besides the standard form of the protein ray-based descriptor, we also consider several other representations of the proteins and make examination which is the most appropriate representation. Some experimental results are given and discussed.
|
M. Riedel (University of Iceland, Reykjavik, Iceland), C. Barakat (Juelich Supercomputing Centre, Juelich, Germany), S. Fritsch (RWTH Aachen Hospital, Aachen, Germany), M. Aach, J. Busch, A. Lintermann, A. Schuppert (Juelich Supercomputing Centre, Juelich, Germany), S. Brynjólfsson, H. Neukirchen, M. Book (University of Iceland, Reykjavik, Iceland)
Enabling Hyperparameter-Tuning of AI Models for Healthcare Using the CoE RAISE Unique AI Framework for HPC
The European Center of Excellence in Exascale Computing "Research on AI- and Simulation-Based Engineering at Exascale" (CoE RAISE) is a project funded by the European Commission. One of its central goals is to develop a Unique AI Framework (UAIF) that simplifies the development of AI models on cutting-edge supercomputers. However, those supercomputers' High-Performance Computing (HPC) environments require the knowledge of many low-level modules that all need to work together in different software versions (e.g., TensorFlow, Python, NCCL, PyTorch) and various concrete supercomputer hardware deployments (e.g., JUWELS, JURECA, DEEP, JUPITER and other EuroHPC Joint Undertaking HPC resources). This paper will describe our analyzed complex challenges for AI researchers using those environments and explain how to overcome them using the UAIF. In addition, it will show the benefits of using the UAIF Hypertuning capability to make AI models better (i.e., better parameters) and faster by using HPC. Also, to demonstrate that the UAIF approach is indeed simple, we describe the adoption of selected UAIF building blocks by healthcare applications. Examples include AI models for Covid-19 chest x-ray analysis and AI models for the Acute Respiratory Distress Syndrome (ARDS). Finally, we highlight other AI models of RAISE use cases that co-designed the UAIF.
|
|
Basic information: Chairs:
Karolj Skala (Croatia), Aleksandra Rashkovska Koceva (Slovenia), Davor Davidović (Croatia)
Steering Committee:
Marian Bubak (Poland), Jesús Carretero Pérez (Spain), Tiziana Ferrari (Netherlands), Dieter Kranzlmüller (Germany), Ludek Matyska (Czech Republic), Dana Petcu (Romania), Uroš Stanič (Slovenia), Matjaž Veselko (Slovenia), Yingwei Wang (Canada)
Program Committee:
Enis Afgan (Croatia), Viktor Avbelj (Slovenia), Davor Davidović (Croatia), Matjaž Depolli (Slovenia), Simeon Grazio (Croatia), Marjan Gusev (North Macedonia), Vojko Jazbinšek (Slovenia), Jurij Matija Kališnik (Germany), Zalika Klemenc-Ketiš (Slovenia), Dragi Kocev (Slovenia), Gregor Kosec (Slovenia), Miklos Kozlovszky (Hungary), Lene Krøl Andersen (Denmark), Željka Mihajlović (Croatia), Panče Panov (Slovenia), Tonka Poplas Susič (Slovenia), Aleksandra Rashkovska Koceva (Slovenia), Karolj Skala (Croatia), Viktor Švigelj (Slovenia), Ivan Tomašić (Sweden), Roman Trobec (Slovenia), Roman Wyrzykowski (Poland)
Registration / Fees:
|
REGISTRATION / FEES
|
Price in EUR
|
EARLY BIRD
Up to 8 May 2023 |
REGULAR
From 9 May 2023 |
| Members of MIPRO and IEEE |
230 |
260 |
| Students (undergraduate and graduate), primary and secondary school teachers |
120 |
140 |
| Others |
250 |
280 |
The discount doesn't apply to PhD students.
Contact:
Karolj Skala
Rudjer Boskovic Institute
Center for Informatics and Computing
Bijenicka 54
HR-10000 Zagreb, Croatia
E-mail: skala@irb.hr
SUBMISSION GUIDELINE:
All submitted papers will pass through a plagiat control and blind peer review process with at least 2 international reviewers.
On the basis of reviewers' opinion and voting result from the conference attendance we will qualify the Best paper for the prize that will be awarded as a part of the final event at the DS-BE conference.
Accepted papers will be published in the ISSN registered conference proceedings. Presented papers in English will be submitted for inclusion in the IEEE Xplore Digital Library.
..........................................................
JOURNAL SPECIAL ISSUE
Authors of the best scientific papers will be invited to submit an extended version of their work to the Scalable Computing: Practice and Experience (ISSN 1895-1767) Journal.
..........................................................
There is a possibility that the selected scientific papers with some further modification and refinement are being published in the following journals: Journal of Computing and Information Technology (CIT), MDPI Applied Science, MDPI Information Journal, Frontiers and EAI Endorsed Transaction on Scalable Information Systems.
Location:
 Opatija is the leading seaside resort of the Eastern Adriatic and one of the most famous tourist destinations on the Mediterranean. With its aristocratic architecture and style, Opatija has been attracting artists, kings, politicians, scientists, sportsmen, as well as business people, bankers and managers for more than 170 years.
The tourist offer in Opatija includes a vast number of hotels, excellent restaurants, entertainment venues, art festivals, superb modern and classical music concerts, beaches and swimming pools – this city satisfies all wishes and demands.
Opatija, the Queen of the Adriatic, is also one of the most prominent congress cities in the Mediterranean, particularly important for its ICT conventions, one of which is MIPRO, which has been held in Opatija since 1979, and has attracted more than a thousand participants from over forty countries. These conventions promote Opatija as one of the most desirable technological, business, educational and scientific centers in South-eastern Europe and the European Union in general.
For more details, please visit www.opatija.hr and visitopatija.com.
|
|
|
|
| Currently there are no news |
|
|
|
|
|