/energy-guided-ot

Primary LanguageJupyter Notebook

🔥 Energy-guided EOT

Learning Energy Based Model guided by Entropic Optimal Transport to learn joint plan $\pi_\theta(x, y)$, which allows to sample jointly close $x$ and $y$ or from conditional projections $\pi_\theta(x | y)$ and $\pi_\theta(y | x)$ as a result of training a single EBM.

Structure

  • theory.ipynb - notebook with theoretical introduction and derivations
  • gauss_swissroll_experiments.ipynb - implementation of experiment with 2D-Gaussian dataset and Swissroll

Visual Results

...

License: MIT