Physics-Informed Video Prediction for Ink Diffusion in Water

Physics-informed prediction with a 28.6% reduction in test MSE over the TAU-only baseline.

May 2026 – July 2026

Course Project, Introduction to Deep Learning. Advisor: Prof. Suyun Zhao.

  • Co-developed a physics-informed video prediction model in a four-member team; handled data processing and implemented the physics-informed prediction module.

  • Reduced test MSE by 28.6% over the TAU-only baseline (280 to 200), achieving 0.9420 SSIM and 32.11 dB PSNR through iterative training and controlled ablation studies.