Optimization of Hybrid Algorithms to Minimize the Loss Function in PINNs

Physics-informed neural networks (PINNs) fold governing equations directly into the loss function, which makes that loss landscape considerably harder to optimise than a standard supervised-learning problem. This project explored hybrid optimisation strategies — combining gradient-based and heuristic search — to train PINNs more reliably, using PyTorch and Optuna for hyperparameter search.

This page is still being filled in with methodology and results — for now, this is the one-line summary from the CV.