Constrained Lagrangian Neural Networks with Learned Physical Constraints

This project is conducted as my Bachelor’s Thesis at Boğaziçi University, co-supervised by Asst. Prof. İnci Meliha Baytaş and Asst. Prof. Arkadaş Özakin, starting in September 2025.

Project Overview

This research extends the framework of Lagrangian Neural Networks (LNNs) by introducing automated constraint learning, forming a Constrained Lagrangian Neural Network (CLNN) architecture.
The model jointly learns the Lagrangian and system constraints from trajectory data, enabling physically consistent modeling even when the governing equations are partially constrained or unknown.