CAT: Interpretable Concept-based Taylor Additive Models

This repository contains the code for the paper CAT: Interpretable Concept-based Taylor Additive Models by Viet Duong, Qiong Wu, Zhengyi Zhou, Hongjue Zhao, Chenxiang Luo, Eric Zavesky, Huaxiu Yao, Huajie Shao.

Model Overview

CAT consists of two main components: concept encoders and Taylor Neural Networks (TaylorNet). Each concept encoder embeds a group of low-level features into a one-dimensional high-level concept representation. The TaylorNet is a white-box model that uses the high-level concept representations to make predictions.

Library Setup

git clone https://github.com/vduong143/CAT-KDD-2024.git
cd CAT-KDD-2024
conda create --name CAT python=3.9
conda activate CAT
pip install -r requirements.txt

Datasets

The datasets used in the paper can be obtained from here. The datasets should be placed in the data folder.

Training and Evaluation

Run CAT training and evaluation on Airbnb:

bash ./scripts/run_airbnb.sh

Run CAT training and evaluation on COMPAS:

bash ./scripts/run_compas.sh

Run CAT training and evaluation on Diabetes:

bash ./scripts/run_diabetes.sh

Run CAT training and evaluation on UCI-HAR:

bash ./scripts/run_har.sh

Run CAT training and evaluation on MNIST:

bash ./scripts/run_MNIST.sh

Run CAT training and evaluation on CelebA:

bash ./scripts/run_CelebA.sh

The gpu variable should be set to the GPU number you want to use. The random seed can be set to desired any integer value. Both variables can be set in the run_*.sh scripts (e.g., run_airbnb.sh).

Interpretability Evaluation

The code for visualizing and interpreting the predictions of the CAT (order 2) model can be found in the notebook CAT_interpretation.ipynb. Please change the data_name variable to the dataset you want to visualize (airbnb or CelebA).

Citation

If you find this repository useful in your research, please cite the following paper:

@inproceedings{duong2024cat,
  title={CAT: Interpretable Concept-based Taylor Additive Models},
  author={Duong, Viet and Wu, Qiong and Zhou, Zhengyi and Zhao, Hongjue and Luo, Chenxiang and Zavesky, Eric and Yao, Huaxiu and Shao, Huajie},
  booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  year={2024}
}

Acknowledgements

This work was supported in part by the AT&T CDO.