Skip to content

KurtButler/2024_attributions_paper

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Explainable Learning with Gaussian Processes

In this repo, we provide MATLAB code to reproduce the results from our paper Explainable Learning with Gaussian Processes, which is currently available on the arXiv.

Abstract: The field of explainable artificial intelligence (XAI) attempts to develop methods that provide insight into how complicated machine learning methods make predictions. Many methods of explanation have focused on the concept of feature attribution, a decomposition of the model's prediction into individual contributions corresponding to each input feature. In this work, we explore the problem of feature attribution in the context of Gaussian process regression (GPR). We take a principled approach to defining attributions under model uncertainty, extending the existing literature. We show that although GPR is a highly flexible and non-parametric approach, we can derive interpretable, closed-form expressions for the feature attributions. When using integrated gradients as an attribution method, we show that the attributions of a GPR model also follow a Gaussian process distribution, which quantifies the uncertainty in attribution arising from uncertainty in the model. We demonstrate, both through theory and experimentation, the versatility and robustness of this approach. We also show that, when applicable, the exact expressions for GPR attributions are both more accurate and less computationally expensive than the approximations currently used in practice.

Instructions

To generate all figures (as .png files), you just need to run main.m. The code should run with no issues using Matlab 2022a or later. All generated figures and tables will be saved to the results folder.

git clone https://github.com/KurtButler/2024_attributions_paper

Data Availability

In our experiments, we used several publicly available data sets from the UCI Machine Learning Repository:

Citation

If you use any code or results from this project in your academic work, please cite our paper:

@article{butler2024explainable,
      title={Explainable Learning with Gaussian Processes}, 
      author={Kurt Butler and Guanchao Feng and Petar M. Djuric},
      year={2024},
      eprint={2403.07072},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Releases

No releases published

Packages

No packages published

Languages