The final project handout is now available.

  • Project handout: AISE2025 Final Project (PDF)
  • Due date: January 26, 2026
  • Submission: one zip file named firstname_secondname_leginumber.zip, including code and a report (max 2 pages per task; figures allowed)
  • Grading: total 140 points + up to 50 bonus points (a grade of 6.0 requires at least 120 points)

Project Overview

The final project consists of three tasks that reflect major topics of the course:

  1. Visualizing Loss Landscapes: PINNs vs Data-Driven (40 points + 20 bonus)
    Compare optimization behavior between PINN and supervised training for a multiscale Poisson problem, and analyze how landscape complexity changes with frequency.

  2. Training an FNO for a Dynamical System (50 points + 10 bonus)
    Train and evaluate Fourier Neural Operators under one-to-one and all-to-all settings, test resolution transfer, and study finetuning under distribution shift.

  3. Geometry-Aware Operator Transformer (GAOT) (50 points + 20 bonus)
    Build a baseline with the official GAOT implementation, then extend it with random sampling tokenization and dynamic-radius aggregation; bonus includes redesigned positional encoding.

Please refer to the PDF for complete task statements, datasets, references, and bonus details.