Authors:
- Carlos Cuevas Villarmin M1-AI
- Jose Felipe Espinosa Orjuela M1-AI
- Javier Alejandro Lopetegui González M1-AI
This repository contains our experiments for the final project of the Bayesian Machine Learning course at ENS-Paris-Saclay.
Title: Review of the paper Loss-Calibrated Approximate Inference in Bayesian Neural Networks.
The code in this repository is mainly based on the implementation of the paper's authors available in the github repository https://github.com/AdamCobb/LCBNN.
Instructions for running the code:
- Install a miniconda distribution compatible with python-3.6 version (23.10 recommended). An archive of miniconda distributions for any Operative system is availabe in miniconda-repo.
- Create a conda environment using the provided environment Bayesian.yml:
conda env create -f bayesian.yml - Activate environment:
conda activate bayesian
Chest X-rays images classification experiments:
The code for the experiments with chest X-rays images for claassifications is based on the experiment done in the paper On Calibrated Model Uncertainty in Deep Learning. The dataset is built from images availables in covid-dataset and kaggle-Chest-X-ray-images.
The code for this experiments is available in the notebooks experiments_x_rays_images_convolutional and experiments_x_rays_images_simple_model.