RealPDE Competition
RealPDE asks how well machine learning models predict real fluid flow. It pairs wind-tunnel measurements (PIV) with matching simulations (CFD) of flow around a NACA4418 airfoil across Reynolds numbers and angles of attack.
We competed in both tracks. Track 1 (Sim2Real) transfers models trained on simulation to real measurements. Track 2 (Long-Term Test-Time Adaptation) forecasts the flow over long horizons while the model adapts online. Every submission was scored on five measures (relative L2 error, turbulent kinetic energy, mean velocity profile error, runtime, and a safe prediction score for calibrated uncertainty) through a combined formula the organizers never published.
Team: Taher Merchant, Aarush Mahajan, Rishi Garg. Advised by Seifedine Kadry. (With thanks to Abhinav Joshi and Siddharth Bharti for their help and contributions in Track 1.)
- Track 2 standing
- Top 20
- Track 2 final score
- 81.82
- Status
- Invited to the Track 2 final evaluation as a reserve team. Final results are announced 10 November 2026.
