NCC Montenegro representative Luka Filipovic took an active part as a Teaching Assistant in the Nvidia and EuroCC2 supported bootcamp “AI for Scientific Computing”, successfully organized by NCCs Germany, Sweden, Austria and Montenegro, on June 25–26, 2024.
The Bootcamp offered a comprehensive introduction to deep neural networks, focusing on applications in scientific computing and physical systems defined by differential equations. The curriculum included advanced topics based on NVIDIA Modulus to develop and train the models in various areas. This online Bootcamp features guided instructions and support from cross-NCC teaching assistants to facilitate learning, supporting participants to build and enhance AI/DL models.
Representatives from NCC Montenegro in joint efforts with young researchers from UDG, published two scientific papers at the SymOrg 2024 conference, organized by the Faculty of Organizational Science, University of Belgrade, at Zlatibor, Serbia on June 12-14, 2024. The conference, traditionally envisioned as a platform for knowledge innovation and empirical research, bringing together representatives from the scientific and professional community, was themed: ”Unlocking The Hidden Potential Of Organization Through Merging Of Humans And Digitals”, aiming to address the newfound need for balance in the era of AI.
Image source: SymOrg 2024 website
The scientific paper “Detection of Scoliosis” by Elvis Taruh, Enisa Trubljanin, and Dejan Babić explores the application of a deep learning model integrated with a web application to detect scoliosis using x-ray images. Utilizing a dataset of 198 x-ray images from Roboflow, the initial model performance was unsatisfactory, prompting manual annotation of 245 images, which significantly improved the model’s accuracy. YOLOv8, a state-of-the-art object detection algorithm, was used to train two models, demonstrating improved performance with manual annotations. The web application, built with Flask, HTML, CSS, and JavaScript, provides a user-friendly interface for analyzing scoliosis detection results. The backend uses MySQL for data storage and management, facilitating efficient image processing, result display, and feedback from doctors. Evaluation metrics indicate that the second model, which underwent refined annotation and augmentation, performed better, avoiding overfitting and demonstrating higher precision. This approach enhances early scoliosis diagnosis and offers a scalable solution for other medical detection challenges, supporting healthcare providers with more accurate diagnostic tools and improving patient care.
Click on image to open SymOrg 2024 proceedings
In the paper “LLM Consistent Character Bias”, the authors Igor Culafic and Tomo Popovic investigate the potential of Large Language Models (LLMs) for character imitation in media, education, and entertainment. Traditionally, LLMs have been used for tasks like web search and programming, but this study focuses on their application in mimicking specific characters from books. Using a dataset created from the Ciaphas Cain anthology of Warhammer 40k, the authors trained models using Low-Rank Adaptation (LoRA) methods. Three models of varying sizes (1.1B, 7B, and 10.7B parameters) were tested, with training conducted on a NVIDIA RTX 4090 GPU. The study found that the larger models (7B and 10.7B) performed well in maintaining character consistency, though they occasionally struggled with specific details and displayed unexpected behaviors like excessive emoji usage. The smallest model (1.1B), despite higher LoRA Rank parameters, was less effective and prone to errors such as repetitive responses and long rants. The authors conclude that LLMs can successfully imitate fictional characters given adequate data and training, suggesting future improvements could make them useful in various fields, including education and therapy. These models have the potential to enhance interactive experiences in theme parks, video games, and educational tools by providing authentic character interactions. However, they caution against using these models as replacements for human therapists due to their limitations and tendency for inaccuracies.
Click on image to open SymOrg 2024 proceedings
Both research papers were partly supported by the EuroCC2 project that is funded by the European High-Performance Computing Joint Undertaking (JU) under Grant Agreement No 101101903.
NCC Montenegro successfully submitted the proposal No. EHPC-BEN-2023B12-015 High-Resolution Weather Prediction Model for Montenegro, to the EuroHPC Benchmark Access Call, in cooperation with Institute of Hydrometeorology and Seismology of Montenegro (IHMS) on VEGA CPU for the period 15.01.-15.04.2024. The project aimed to leverage the EuroHPC resources to establish and benchmark precise weather forecast models in the complex topography of Montenegro, utilize these models to refine existing meteorological models, and ultimately enhance the accuracy of weather forecasts, particularly for severe weather events.
Simulations used the Weather Research and Forecasting Non-Hydrostatic Mesoscale Model (WRF-NMM) combining advanced numerical techniques with HPC, for studying atmospheric phenomena with high spatial and temporal resolution and providing accurate and efficient simulations of regional weather patterns. Key activities included installing and fine-tuning the model based on previous verification results, preparing input data, running and fine-tuning the model, and analyzing results in the context of weather prediction and parallel computing performance. WRF model is tested on complex Montenegrin terrain on resolutions 0.5km, 1km, 3km, and 5km. Application scalability was tested on up to 8 nodes, running up to 1024 tasks simultaneously. Simulations were executed for different timespans, but results/overall execution time was scaled to one day period to calculate application speedup and efficiency.
Vega HPC has significantly enhanced research capabilities, allowing to achieve results more quickly and with greater accuracy: scalability was successfully tested on 64-512 CPU cores and the model was successfully downscaled to the resolution of 0.5 km. The final report on granted EUROHPC JU Benchmark Access and effective utilization of the VEGA CPU system assigned is submitted.
PAID-T (Price Action Intelligent Detection Trading) is a scalable software solution for investment funds, banks, brokers and digital banks that automatically analyzes markets, executes trades and optimizes trading. The basis of the solution is the awareness of the context when the most potent moment for trading is, the ability to quickly adapt to new market conditions, as well as rigorous rules for risk management.
PAID-T uses mathematical models for statistical data processing (big data), instrument price detection algorithms, machine learning, artificial intelligence and blockchain technology to create a unique trading experience.
Using the Lumi supercomputer will help us significantly speed up historical testing. Before using Lumi resources, our tests took an average of 6 days. We now expect to be able to complete testing between 5 and 12 hours depending on the number of parameters used.
The PAID-T has been supported by NCC Montenegro providing technical consultation, HPC systems expertise, and support regarding the application process for EuroHPC Development Call on LUMI (CPU partition). HPC infrastructure access was approved and provided for 12 months.
Representative of NCC Montenegro Ms. Sanja Nikolic participated in the HPC Masterclass, a hybrid event held at “Ovidius” University of Constanța (20-24.05.2024) co-organized by NCC Romania. The second day of the training event was dedicated to NCC’s shared experience on HPC academic utilization and research excellence, introducing presenters from NCC Portugal, Montenegro, Bulgaria, France, and Luxembourg. Ms. Nikolic presented the development of the HPC+ (HPC and related technologies) educational ecosystem in Montenegro, intensively driven by the EuroCC project, NCC Montenegro and the University of Donja Gorica.
NCC Montenegro is focused on developing HPC/HPDA/AI academic programs, study courses, and training portfolios, following a sustained learning pathway:
• HPC/AI-related seasonal schools to mobilize high-school students and high-tech enthusiasts to enroll in HPC/AI-related academic programs. (“Open Mind Academy”). • BSc restructured program (M1-Software Development and M2-Digital Transformation), and first nationally accredited AI Master Program, at Faculty for Information Systems and Technologies. • Professional training courses, workshops, and networking events for academia and industry participants providing in-demand HPC/AI knowledge, technical upskilling, and supercomputing hands-on sessions.
NCC Montenegro’s training portfolio covers technology-specific know-how (HPC system architecture and applications, Parallel Programming, Python Programming, Deep Learning, Edge IoT, Computer Vision, NLL) and industry-specific priority domains, defined by the Smart Specialization Strategy of Montenegro (energy, health, tourism, agriculture, ICT).
NCC Montenegro is also fully exploiting joint training opportunities within the pan-European NCC network – to capitalize on their HPC expertise and resources, as well as productive collaboration with key MNE stakeholders (business associations, technical affiliations, and funding institutions) – to enhance national HPC/AI awareness, outreach and uptake.
The AI for Scientific Computing Bootcamp provides a step-by-step overview of the fundamentals of deep neural networks and walks attendees through the hands-on experience of building and improving deep learning models for applications related to scientific computing and physical systems defined by differential equations. The material will cover more advanced topics such as physics-informed neural networks (PINNs) and operator learning and make use of tools like NVIDIA Modulus to develop and train the models. This online bootcamp is a hands-on learning experience where you will be guided through step-by-step instructions with teaching assistants on hand to help throughout.
NCC Montenegro, in collaboration with Montenegrin National Committee of CIGRE, one of the leading worldwide organizations on electric power systems, successfully organized a collaborative workshop, focusing on opportunities and benefits of HPC and AI complementary technologies in the energy sector, industry best practices and examples of high-fidelity energy metrics modelling. The workshop was attended by approx. 30 participants, academic researchers and industry experts, effectively converging HPC-powered AI research advancements with challenges and opportunities in the energy sector, driven by quality of predictive analytics and sustainability of energy models.
Representatives of HPC National Competence Centre of Montenegro, Sanja Nikolic and Luka Filipovic presented EUROCC2 project, NCC services, EuroHPC infractructure acces, PC4SME assessment tool and research, as well as success stories demonstrating advancements in HPC and AI applications in energy sector.
Mr Darko Krivokapić, Executive Director of the Directorate for Energy Management of the Montenegrin Electric Enterprise (EPCG), presented characteristics of Montenegro energy system and challenges of short- and long-term predictions of energy prices and consumption, depending on Montenegrin energy balance, renewable energy sources, weather conditions and trade dynamics/external risks facing electricity wholesale markets.
Mr Milutin Pavicevic PhD student@University of Donja Gorica and CEO of Alicorn, presented forecasting day-ahead electricity metrics (prices and loads) with different models of artificial neural networks and its accuracy comparison on the given dataset, elaborating on possible commercial implications in energy sector.
Mr Lazar Scekic, Teaching Assistant @University of Montenegro, introduced the audience with its research work on security challenges of electricity infrastructure (GPS spoofing) and reliable AI methods for the protection of electrical power systems against cyberattacks.
Mr Ivan Vujovic, PhD student @University of Belgrade and CEO of Tering, presented models related to prediction of production and consumption of electrical energy in the electrical system by using recurrent neural networks, discussing quality parameters that determine and improve predictive functioning of energy systems.
HPC and AI energy workshop enhanced awareness of NCC Montenegro services and EuroHPC infrastructure resources, available free of charge for the Montenegrin industry for the purpose of research, development and innovation.
Intensive discussion and exchange of opinions, expertise and experience between energy systems’ researchers and practitioners, including meteorological professionals also, open possibilities for the productive collaboration on calibrating computationally-demanding AI/ML models for national energy datasets, to improve their learning accuracy, prediction rates and potential of its commercial utilization in the energy system of Montenegro.
