Edin Jelačić, Doctoral student
Education Background:
Ph.D. Researcher in Computer Science, Mälardalen University, Sweden, 2023 – Present.
Master’s Degree in Robotics, University of Ljubljana, Slovenia, 2019 – 2021.
Bachelor’s Degree in Automatic Control and Electronics, Sarajevo University, Bosnia and Herzegovina, 2016 – 2019.
Work: Prior to his doctoral studies, Edin worked as a Software Engineer at the Control Systems Laboratory at Cosylab in Slovenia from 2019 to 2023. His industry work focused heavily on the engineering of critical software components for high-level radiation therapy control systems. He has hands-on experience architecting safety analysis frameworks to ensure regulatory compliance for clinical facilities, implementing SCADA protocols, and executing on-site engineering for international medical clients. His background combines high-performance engineering (C++, C#, Java) with real-world, safety-critical deployments.
Research: Edin’s research focuses on the intersection of Safety-Critical Systems, Verifiable AI, and Embedded Intelligence. He is currently a fully-funded PhD Researcher under the Knowledge Foundation's "PerFlex" project. His research is strongly problem-oriented and conducted in close collaboration with industrial partners such as Volvo Cars, Hitachi Energy, and Ericsson.
His goal is to bridge the gap between advanced artificial intelligence and strict industrial safety standards. His recent work includes architecting automated vehicular scenario generation pipelines for autonomous driving validation, developing deep-learning-based cache simulators, and engineering CPU load forecasting frameworks that utilize Conformal Prediction and Shapley values for real-time interpretability on constrained hardware.
His research interests include, but are not limited to:
Verifiable AI and Machine Learning Reliability
Safety-Critical Embedded Systems and Control
Automated Validation and Simulation for Autonomous Vehicles
Deep Learning, Conformal Prediction, and Interpretability
Tools I use: Python, C++, Java, PyTorch, TensorFlow, Docker, MATLAB/Simulink, esmini, Electron.js.
Positions:
Knowledge Foundation (KK-Stiftelsen) PerFlex Project, Researcher, 2023 – present.
Teaching & Supervision: Edin is actively involved in academic mentorship. He supervises multiple Master's thesis students, guiding their research in AI, Reinforcement Learning, and the application of verifiable intelligence to engineering problems. Teaching assistant for DVA245 and DVA258.
Select Publications:
"A Conformal Prediction-Based Framework for CPU Load Forecasting: A Black-Box Approach", COMPSAC, 2025.
"Machine-Learning Based Cache Miss Prediction", Springer International Journal on Software Tools for Technology Transfer (STTT), 2025.
"Abstraction-based Reduction of Input Size for Neural Networks", AISoLa, 2023.
Research topic: Safety-critical systems and autonomous environments require not just high performance, but absolute reliability. My research mainly focuses on closing the gap between advanced, high-performance artificial intelligence and strict industrial safety standards. Through the Knowledge Foundation's "PerFlex" project (Performant and Flexible digital Systems through Verifiable AI), I am investigating methods to guarantee the safety and reliability of embedded intelligence.
I am actively working on developing automated validation pipelines and simulation environments for autonomous driving features under industrial safety constraints. Furthermore, my research explores resource optimization and predictability in embedded hardware, utilizing techniques like Conformal Prediction for uncertainty quantification, Shapley values for real-time interpretability, and deep learning models for CPU load and cache miss forecasting. My goal is to ensure that AI deployed in safety-critical domains is both highly performant and rigorously verifiable.
Tools: HASCo (Hybrid AI Scene Compiler) integrated with the esmini simulation environment for orchestrating complex, automated vehicular traffic scenarios for autonomous drive validation. Marabou-based NN Reduction: Co-authored an algorithm extending the formal verification tool Marabou to perform abstraction-based reduction of input sizes for neural networks. AI Deployment App: Architected a full-stack Electron/Typescript application with Docker containerization to streamline AI model deployment and ensure reproducible testing across simulation environments. Deep Learning Cache Simulator: Engineered a simulator using LSTM networks to predict cache miss distributions with near-native accuracy.
Recent experience:
2025 - I publish "A Conformal Prediction-Based Framework for CPU Load Forecasting: A Black-Box Approach" at the COMPSAC 2025 Conference in Toronto, Canada.
2025 - I publish "Machine-Learning Based Cache Miss Prediction" in the Springer International Journal on Software Tools for Technology Transfer (STTT), Germany.
2023 - I give a presentation regarding the "Abstraction-based Reduction of Input Size for Neural Networks" at the AISoLa 2023 Conference in Greece.
2023 - I begin my fully-funded PhD research under the PerFlex project at Mälardalen University, collaborating with industrial partners such as Volvo Cars, Hitachi Energy, and Ericsson.
2021 - 2023 - I work as a Software Engineer at Cosylab, architecting unit safety analysis frameworks and engineering critical Java modules for radiation therapy control systems, including on-site engineering in Helsinki, Finland.
2021 - I present my Master's Thesis, "Dynamic Movement Primitives in High-Precision Robotic Applications," at the 30th International Electrotechnical and Computer Science Conference (ERK 2021).
A Conformal Prediction-Based Framework for CPU Load Forecasting: A Black-Box Approach (Jul 2025) Edin Jelačić, Cristina Seceleanu, Peter Backeman, Ning Xiong, Tiberiu Seceleanu, Axel Jantsch 49th IEEE International Conference on Computers, Software, and Applications ( COMPSAC-2025)
Machine learning-based cache miss prediction Edin Jelačić, Cristina Seceleanu, Ning Xiong, Peter Backeman, Sharifeh Yaghoobi , Tiberiu Seceleanu International Journal on Software Tools for Technology Transfer (STTT)