Annealed and Transformed Variational Inference

Official code for [Fast Computer Model Calibration using Annealed and Transformed Variational Inference]

ATVI adopts surjective function and temperature annealing to avoid inferential challenges in a naive VI approach in calibration tasks.

This code relies on the NF model package from the SurVAE library. To start with, you must install the SurVAE library.

Installation

In the folder containing setup.py, run

pip install .

Experiments

To train or evaluate model, please check /experiments.

Paper

The original paper can be found here: *https://www.tandfonline.com/doi/full/10.1080/10618600.2024.2374962

Acknowledgements

I would like to thank the authors who provide useful repositories. Mainly,

• https://github.com/didriknielsen/survae_flows
• https://github.com/rtqichen/torchdiffeq