ENCA-INCA

Learning Summary Statistics for Bayesian Inference with Autoencoders

The code here implements the work proposed by Albert, C., Ulzega, S., Ozdemir, F., Perez-Cruz, F. and Mira, A. (2022). Learning Summary Statistics for Bayesian Inference with Autoencoders. arXiv preprint arXiv:2201.12059.

Contents

The repository contains all the code needed to reproduce the proposed models explicit noise conditional autoencoder (ENCA) and implicit noise conditional autoencoder (INCA) for encoding near-sufficient and highly concentrated summary statistics for the two statistical models experimented in this work.

Demo

The repository can be set up on a clean environment by creating a conda environment by

conda env create -f environment.yml
conda activate encainca

In order to train models ENCA and INCA for statistical model 1 and 2, one can use the provided scripts:

python train_ENCA_model1.py
python train_INCA_model1.py
python train_ENCA_model2.py
python train_INCA_model2.py

Interactive Environment

The project can also be run on renku environment. This allows to skip the step requiring to set up an environment to test the repo.

Prerequisites

Code is tested for Tensorflow v2.4.1. However preliminary experiments suggest it should also work on TF v2.2.
Please report any issues if you come across bugs.

Citation

If you use any content of this repository, please use the following bibtex:

@article{albert2022learning,
  title={Learning Summary Statistics for Bayesian Inference with Autoencoders},
  author={Albert, Carlo and Ulzega, Simone and Ozdemir, Firat and Perez-Cruz, Fernando and Mira, Antonietta},
  journal={arXiv preprint arXiv:2201.12059},
  year={2022}
}