Towards a Unified Framework for Evaluating Explanations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pinto, Juan D., Paquette, Luc
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917721354010624
author Pinto, Juan D.
Paquette, Luc
author_facet Pinto, Juan D.
Paquette, Luc
contents The challenge of creating interpretable models has been taken up by two main research communities: ML researchers primarily focused on lower-level explainability methods that suit the needs of engineers, and HCI researchers who have more heavily emphasized user-centered approaches often based on participatory design methods. This paper reviews how these communities have evaluated interpretability, identifying overlaps and semantic misalignments. We propose moving towards a unified framework of evaluation criteria and lay the groundwork for such a framework by articulating the relationships between existing criteria. We argue that explanations serve as mediators between models and stakeholders, whether for intrinsically interpretable models or opaque black-box models analyzed via post-hoc techniques. We further argue that useful explanations require both faithfulness and intelligibility. Explanation plausibility is a prerequisite for intelligibility, while stability is a prerequisite for explanation faithfulness. We illustrate these criteria, as well as specific evaluation methods, using examples from an ongoing study of an interpretable neural network for predicting a particular learner behavior.
format Preprint
id arxiv_https___arxiv_org_abs_2405_14016
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Towards a Unified Framework for Evaluating Explanations
Pinto, Juan D.
Paquette, Luc
Machine Learning
Artificial Intelligence
The challenge of creating interpretable models has been taken up by two main research communities: ML researchers primarily focused on lower-level explainability methods that suit the needs of engineers, and HCI researchers who have more heavily emphasized user-centered approaches often based on participatory design methods. This paper reviews how these communities have evaluated interpretability, identifying overlaps and semantic misalignments. We propose moving towards a unified framework of evaluation criteria and lay the groundwork for such a framework by articulating the relationships between existing criteria. We argue that explanations serve as mediators between models and stakeholders, whether for intrinsically interpretable models or opaque black-box models analyzed via post-hoc techniques. We further argue that useful explanations require both faithfulness and intelligibility. Explanation plausibility is a prerequisite for intelligibility, while stability is a prerequisite for explanation faithfulness. We illustrate these criteria, as well as specific evaluation methods, using examples from an ongoing study of an interpretable neural network for predicting a particular learner behavior.
title Towards a Unified Framework for Evaluating Explanations
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2405.14016