Mr. Veselin Andric defended his BSc thesis titleld “Prompt endineering for LLMs” at the Faculty for information systems and technologies. The devence took place on 2 Oct 2024 and it was done under mentorship of the EuroCC and HPC4S3ME teams’ members. This was a part of the effort to promote HPC and AI related technologies in the teaching curricula and research activities at UDG.
ABSTRACT – Prompt engineering is one of the primary areas of Natural language processing (NLP). It is a process that involves designing and improving inputs that are given to a language model such as ChatGPT, with a goal of getting wanted results. This dissertation investigates details of prompt engineering, it’s theoretical foundation, methodologies and practical uses in different tasks of NLP.
Mr. Veselin Andric defended his BSc thesis on Prompt EngineeringThe provided an overview of NLP, LLMs and prompt techniques
At the Avala Hotel in Budva, on September 27, 2024, as part of the international EPhEU event, the association of pharmacy organizations from Europe (September 26 to 28, 2024), a session on Module 6 of continuing education for Montenegrin pharmacists was held, titled “Artificial Intelligence in Pharmacy: Digital Challenges, Myths and Misconceptions Incorporated in Digital Media, the Need for New Competencies for Pharmacists.” This event gathered over 150 experts, both online and in person, from the field of pharmacy, aiming to discuss the role of artificial intelligence in transforming the pharmaceutical sector.
Mr Cakic gave a lecture on HPC and AI and applications in Pharmaceutical industry
Mr Stevan Čakić from UDG and NCC Montenegro gave a lecture on the history and development of AI, emphasizing its growing importance in pharmacy. Then a special attention was given to the application of AI in drug development, personalized therapy, and optimization of pharmaceutical services. The discussion addressed the potential of AI and HPC to reshape the way pharmacists perform their daily tasks, as well as the need to improve skills and knowledge to adapt to new technologies. Link for the PKFE: https://pkfe.me/edukativni-programi/
On Friday, September 27, UDG welcomed Divya Siddarth, one of the 100 most influential young people in the field of artificial intelligence, as chosen by TIME magazine!
Ms Divya Siddarth is one of the 100 most influential young people in the field of artificial intelligence
Divya is a researcher at Microsoft Research, in the Political Economy and Social Technologies (PEST) department, and co-founder of the Collective Intelligence Project. Her work focuses on using technology to create fairer societies, with a particular emphasis on decentralized technologies, artificial intelligence and digital democracy.
She was recently recognized as one of the key minds shaping the future of artificial intelligence in the service of the common good!
It was a pleasure to be a part of this discussion. There was over 200 people in the audienceDivya is a researcher at Microsoft Research and co-founder of the Collective Intelligence Project
At the “Revolutionising Agriculture with HPC and AI: Real-World Applications” conference, the AIMHiGH project, coordinated by DigitalSmart, was presented as part of the HPC Fortissimo initiative under the Horizon 2020 FF4EuroHPC program. The project, titled “AI/ML Enabled by HPC for Edge Camera Devices for the Next Generation Hen Farms,” brought cutting-edge HPC and AI/ML technologies to the poultry farming sector. Key achievements included real-time monitoring and data-driven insights through AI-powered edge cameras that continuously surveilled poultry, detecting early signs of illness and abnormal behavior. This innovation allowed farmers to intervene promptly, improving animal welfare, reducing losses, and enhancing productivity.
S. Cakic presented the project AIMHiGH that utilised HPC and AI for computer vision in smart agriculture
Additionally, the AI models developed during the project provided advanced solutions for early disease detection, resource optimization, and automation, leading to more sustainable and efficient farming practices. The system reduced manual supervision needs, cut operational costs, and optimized feed and water usage based on real-time data. The technology’s scalability and flexibility made it accessible to both large and small farms. Collaboration with partners like DunavNET, the University of Donja Gorica, and Montenegrin companies Meso-promet Franca and Radinović ensured practical relevance, aligning the project with Montenegro’s Smart Specialisation Strategy while supporting innovation and local economic growth in the poultry sector.
Presenting HPC and deep learning approach to the problemDiscussing the experiment results and prospects of the future wok
AIFusion! Apply for a course on artificial intelligence (AI) in agriculture, medicine and energy. HPC4S3ME team mebers will actively participate in training and workshop sessions. The training is carried out in the context of NCC Montenegro and EUROCC with the financial support of the Innovation Fund of Montenegro. We continue last year’s success! This is an activity supported by the Innovation Fund of Montenegro.
Would you like to explore the transformative power of HPC and Artificial Intelligence in agriculture and learn about real-case application?
Join the webinar “Revolutionising Agriculture with HPC and AI: Real-World Applications on Tuesday, 24 September, 14:00-15:00 CET.
More info & registration ▶ https://www.sling.si/en/news/ncc-sling-webinar-revolutionising-agriculture-with-hpc-and-ai-real-world-applications/
HPC4S3ME team members will participate in the webinar and also present their experiences in developing computer vision solutions for precision farming.
Our young researchers Ms. Zoja Scekic and Ms Tamara Pavlovic are wrapping up their master theses submission and the defence will be scheduled for October. These efforts are mentored and driven though the HPC4S3ME project and represent the main outputs of the project. One of the theses focuses on the HPC/AI applications in energy sector, while the other is focused on applications in medicine, both sectorial priorities of Montenegrin S3 .
Two young female researchers to defend their MSc theses in October (image: ChatGPT)
The first master’s thesis examines the use of advanced deep learning models for day-ahead electricity price prediction, comparing their accuracy and efficiency with traditional methods. With the increasing integration of renewable energy and the complexity of electricity markets, accurate forecasting is essential. The research includes four case studies using different techniques: Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and hybrid CNN-LSTM models. Despite promising results, limitations regarding data quality, model complexity, and computational demands are acknowledged. The study highlights the need for further optimization and broader applications across energy markets.
The second master’s thesis explores the use of AI in breast cancer diagnosis, utilizing computer vision algorithms to analyze mammographic images. By applying convolutional neural networks (CNNs) like ResNet152 and DenseNet121, the study demonstrates how AI can improve early detection, streamline screening processes, and support more personalized treatment approaches. With AUC scores surpassing 0.9, the models show strong potential for clinical use. The thesis also addresses ethical considerations, including patient safety and AI transparency, while emphasizing the need for further research in AI-driven medical diagnostics.
These efforts are mentored and driven though the HPC4S3ME project
The in-house HPC lab at UDG is getting an upgrade. Though cross-project collaboration with the support from EUROCC (NCC Montenegro) and financial support through AI-AGE project financed by the Ministry of education, science and innovation, UDG is procurring additional computing node and data storage that will be integrated with the equipment lab established through HPC4S3ME.
More in-house computing and storage power
The HPC Rack computing node is a high-performance 2U rack server designed for demanding computing tasks. It features dual processors, each with 24 cores running at 3.40 GHz, offering 48 threads and a 36 MB cache. The server comes equipped with 256GB DDR4 memory, expandable up to 4TB per processor, and includes 32 memory slots. Storage is provided by two 480GB SSD SATA 6G drives, with graphics powered by an NVIDIA L40 48GB card. It includes a Broadcom MR216i-a storage controller and network capabilities with 4x 1Gb Ethernet ports and 1 management port. The server also supports advanced security features like UEFI Secure Boot, Trusted Platform Module (TPM) 1.2, and secure firmware updates, ensuring tamper-free operation and data protection. Additionally, it is equipped with dual 1600W power supplies and easy rack-mounting options for streamlined deployment in HPC environments.
The NAS storage system features a robust setup designed for high-capacity data management and sharing. It includes a 4-bay configuration, each supporting 12TB SATA 6Gb/s drives, allowing for a total storage capacity of up to 66TB. Powered by an Intel Celeron N5095 4-core processor with 8GB RAM, this system provides solid performance for various storage needs. With dual 2.5 Gigabit Ethernet ports and PCIe Gen 3 expansion capabilities, it ensures fast data transfer and network connectivity. Additionally, it supports up to 1,500 concurrent CIFS connections, making it suitable for medium to large-scale data environments.
We had a very successful meeting with the representatives from CFCU and Ministry of Education. This meeting was organized in a form of an On-the-spot-verification in accordance to the general agreement of the grant contract. The HPC4S3ME team gave a comprehensive presentation of the project objectives, current implementation status and KPIs, as well as the directions and plan for the remaining time of the project. The visit took place on 14 May 2024.
On-the-spot-verification visit at the UDGHPC4S3ME team hosted representatives from the CFCUWe went over the project implementation details
Our team presented the paper “Ovarian Cancer Detection Using Computer Vision” that addresses the use of AI and HPC tools for computer vision application in medicine. The paper was authored bz A. Abazovic, A. Lekic, I. Jovovic, S. Cakic and T. Popovic. This paper is a result of engaging young researchers and interns to embrace HPC and AI technology. The successful presentation of the paper is followed by its publication in the IEEE Xplore electronics library. Information about the conference is available at the conference website.
Click on image to open
ABSTRACT – This study explores the application of artificial intelligence (AI) and deep learning in the field of computer vision, specifically for the detection of ovarian cancer. A computer vision model was developed, utilizing two different AI models, YOLOv8 and YOLOv7, to evaluate their effectiveness in this medical context. YOLOv8, being the current state-of-the-art model in computer vision, was chosen for its advanced capabilities, while YOLOv7 was selected for its established usage and performance record. Comparative analysis revealed that YOLOv8 outperformed YOLOv7 with a significantly higher accuracy rate of approximately 0.9. This enhanced accuracy is crucial in medical applications, particularly for early cancer detection which can substantially improve patient outcomes. Additionally, the model was benchmarked against other machine learning models and existing computer vision approaches in ovarian cancer detection. While this model demonstrated superior accuracy compared to other machine learning techniques, it was observed that certain other computer vision models, leveraging more customized architectures and larger datasets, achieved marginally better results. These findings indicate potential areas for future improvement of implemented model, including the integration of more comprehensive datasets and the refinement of model architecture. Furthermore, the research proposes the incorporation of additional health parameters to enhance the model’s effectiveness and applicability in medical diagnostics.
Link to the paper at IEEE Xplore is available here.
