Wasserstein Adversarial Generative Network for Optimal Transport
Designed and programmed a Wasserstein Generative Adversarial Network (WGAN) that trains from the MNIST dataset and maps Gaussian noise to realistic handwritten digits. PyTorch and NumPy were used with GPU-friendly integrations.
Neural ODEs for continuous optimal transport
Implemented a neural ODE in Python to solve the well-known Benamou–Brenier optimal transport problem. The program uses an RK4 solver together with classic PyTorch packages. The Gaussian distribution is smoothly transported to the desired digit.
