Toggle Main Menu Toggle Search

Open Access padlockePrints

Challenges of Using XAI for Geographic Data Analytics

Lookup NU author(s): Dr Jin XingORCiD

Downloads


Licence

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).


Abstract

Although eXplainable Artificial Intelligence (XAI) has huge potential to glassbox deep learning models, there are challenges in applying it in the domain of Geospatial Artificial Intelligence (GeoAI), specifically Deep Neural Networks. We summarize these challenges, which include the difficulty of selecting reference data/models, the shortcomings of gradients as explanation, the challenges of accommodating geographic scale, the limitations of knowledge scope in the explanation process of GeoAI, the lack of acknowledging non-technical aspects in XAI, the incompatibility of geography in XAI visualization, as well as underlying geographic data analytical processes that are not amenable to XAI.


Publication metadata

Author(s): Xing J, Sieber R

Editor(s): Mai, G; Cai, L;

Publication type: Conference Proceedings (inc. Abstract)

Publication status: Published

Conference Name: The 1st International Workshop on Methods, Models, and Resources for Geospatial Knowledge Graphs and GeoA

Year of Conference: 2021

Print publication date: 27/09/2021

Online publication date: 27/09/2021

Acceptance date: 31/08/2021

Date deposited: 14/12/2021

URL: https://ling-cai.github.io/GIScience-GeoKG/


Share