Projects
Loss Landscape Exploration
Authors: Joe Koszut, Theodoros Xenakis
Date: December 2025
This project investigates the structure of the loss landscape between modes of deep learning models, with the goal of better understanding training dynamics and improving model robustness. The work was submitted as a semester project for 6.7960 – Deep Learning at MIT and received a final grade of A.
PDE Simulations
Authors: Henrik S. Grønlund, Magnus U. Rønneseth, Theodoros Xenakis
This project was developed as part of TMA4212 at NTNU and received a final grade of A. It explores numerical methods for differential equations through two independent implementations.
Project 1 applies Finite Difference Methods to the SIR epidemiological model, implemented in Julia using Jupyter notebooks. Project 2 studies Finite Element Methods in Python, with applications to the Poisson equation and an optimal control problem for temperature regulation in a rod.
AIS Vessel Position Prediction
Authors: Einar J. Rye, Henrik S. Grønlund, Theodoros Xenakis
This project was developed as part of TDT4173 – Modern Machine Learning in Practice at NTNU and received a final grade of A. The goal was to predict future vessel positions using large-scale AIS time-series data.
The project centers around two standalone machine learning pipelines implemented in Python. Pipeline 1 uses a Random Forest model, while Pipeline 2 applies XGBoost. Both notebooks independently handle preprocessing, feature engineering, model training, postprocessing, and prediction generation.
Scientific Computing Projects
Authors: Henrik S. Grønlund, Magnus U. Rønneseth, Theodoros Xenakis
This project collection was developed as part of TMA4320 – Introduction to Scientific Computing at NTNU and received a final grade of A. It consists of three independent computational projects implemented in Python.
Project 1 explores Monte Carlo simulation of polymer folding on a 2D lattice. Project 2 applies transformer-based neural networks to sequence prediction on synthetic data. Project 3 studies particle transport driven by ocean currents using numerical time integration.
Together, the projects cover stochastic simulation, numerical modeling, and modern machine learning, with emphasis on implementing algorithms from first principles and analyzing their behavior through visualization and experiments.