This repository contains the source code of the paper entitled:

Leveraging PAC-Bayes Theory and Gibbs Distributions for Generalization Bounds with Complexity Measures
Paul Viallard, Rémi Emonet, Amaury Habrard, Emilie Morvant, Valentina Zantedeschi
AISTATS, 2024

Running the experiments

To reproduce the experiments, you have to execute the following commands in your bash shell.

Generating the data:

python run.py local generate_data.ini

Learning the models to generate the dataset for the neural complexity:

python run.py local learn_data_neural_comp.ini
python run.py local merge_data_neural_comp.ini

Learning the neural complexity:

python run.py local learn_neural_comp.ini

Computing the bounds with complexity measures:

python run.py local learn_comp_fig_1.ini
python run.py local learn_comp_fig_2.ini
python run.py local learn_comp_fig_3.ini

Generating the plots:

python run.py local generate_plot.ini

Conda environment

The code was tested with the conda environment in the env.yml.