On 12 December 2025, the University of Montenegro, as a member of the NCC Montenegro (EuroCC) team, successfully organized a short training titled “Building a Neural Network, Code Preparing for Multi-GPU HPC and Running Large-Scale Training” at the Faculty of Science and Mathematics.
The training gathered around 30 participants, including students, young researchers, and industry professionals, all interested in learning how to effectively use High-Performance Computing (HPC) resources in their research and development work.
The course combined theory and hands-on practice through a concrete use case. After an introduction by Prof. Enis Kočan, Prof. Igor Jovančević, together with his colleagues Velibor Došljak and Nikola Pižurica, led an informative, hands-on technical session. Participants first learned how to build a simple neural network for vision-based defect detection in manufacturing, with the goal of deploying it on an edge device with limited computational power. Optimizing the neural network requires many experiments, clearly demonstrating the need for HPC resources.
The focus then shifted to preparing a local development environment using Docker, transferring the environment to an HPC system, and finally running multiple training experiments on a multi-GPU HPC infrastructure.
Throughout the session, participants actively engaged in questions and discussions, particularly regarding the practical and technical aspects of applying for and using HPC resources. The training inspired participants to plan several applications to open calls for HPC resources, particularly in the context of master’s theses and startup projects.
An important takeaway from the training was that the EuroCC2 / NCC Montenegro team provides continuous support — not only during the application process for HPC resources, but also during their actual usage, including environment setup, code adaptation, and performance considerations. This event confirmed strong interest in HPC and AI-related topics and highlighted the importance of hands-on, application-oriented training in bridging the gap between local development and large-scale HPC execution.