/RMwGGIS

Official implementation of "Gradient-Guided Importance Sampling for Learning Binary Energy-Based Models" [ICLR2023]

Primary LanguagePython

Gradient-Guided Importance Sampling for Learning Binary Energy-Based Models

This is the official implementation of the RMwGGIS method proposed in the following paper.

Meng Liu, Haoran Liu, and Shuiwang Ji. "Gradient-Guided Importance Sampling for Learning Binary Energy-Based Models". [ICLR 2023]


Visualization of learned energy functions on 32-dimensional synthetic discrete datasets.

There is an implementation from the community as well.

Requirements

We include key dependencies below.

  • PyTorch
  • tqdm
  • sympy
  • distutils

Run

To run the experiments on synthetic discrete data, please refer to the commands in run.sh.

Reference

@inproceedings{liu2023rmwggis,
  title={Gradient-Guided Importance Sampling for Learning Binary Energy-Based Models},
  author={Liu, Meng and Liu, Haoran and Ji, Shuiwang},
  booktitle={International Conference on Learning Representations},
  year={2023}
}