Concept Bottleneck Model with Emergent Communication Framework

This repository provides code for training a Concept Bottleneck Model integrated with an emergent communication framework using reinforcement learning (PPO). The project is associated with the research paper:
"Concept Bottleneck Model with Emergent Communication Framework for Explainable AI"
by Farnoosh Javar and Kei Wakabayashi,
accepted for publication in the xAI-2025 Late-breaking Work, Demos and Doctoral Consortium Joint Proceedings (published by CEUR-WS),
and to be presented as a poster at The 3rd World Conference on eXplainable Artificial Intelligence (XAI-2025).

Installation

Install the required packages:

pip install -r requirements.txt

Usage

To start training, run:

python main.py

Paths to datasets and training parameters can be adjusted in src/config.py.

The training script expects:

• Pre-extracted feature files (.npz) for train/val/test sets
• HOC annotation CSV file

If needed, preprocessing scripts are provided to generate these files.

Project Structure

main.py
requirements.txt
LICENSE         # <-- MIT License for code
src/
  ├── config.py
  ├── models.py
  ├── environment.py
  ├── utils.py
  ├── train.py
  └── extract_resnet_features.py
Data/
  ├── Generate_Subset.py
  ├── HOC_annotations.csv
  ├── HOC_list.txt              
  ├── HOC_list.txt
  ├── LICENSE    # <-- CC BY-SA 4.0 License for dataset
  └── README.md # <-- dataset-specific README

License

This project is licensed under the MIT License. See the LICENSE file for details.