Multi-Task Symbolic Regression for Gravitational Wave Law Discovery

This project was conducted as a research internship at Radboud University, in the High Energy Physics group under the supervision of Prof. Sascha Caron, Nijmegen, Netherlands, from June 2025 to September 2025.

---

Project Overview

This research explored multi-task symbolic regression using transformer architectures to uncover interpretable physical relations in gravitational wave data.
The approach leveraged contrastive learning to evaluate the likelihood of observed numerical data under candidate symbolic expressions using transformers.

---

Research Goals

• Design transformer-based models capable of discovering mathematical laws from noisy astrophysical data
• Integrate symbolic regression and contrastive learning for interpretable physical inference
• Advance methods for AI-driven theory discovery in high-energy physics

---

Key Technologies

• Transformers: Sequence modeling for symbolic expression generation
• Contrastive Learning: Similarity-based training for expression likelihood estimation
• Symbolic Regression: Interpretable model discovery via expression trees
• Python: Implementation and experimentation environment
• Gravitational Wave Analysis: Data preprocessing and simulation for physical signal modeling