Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ensured: Explanations for Decreasing the Epistemic Uncertainty in Predictions

Helena Löfström, Tuwe Löfström, Johan Hallberg Szabadvary

TL;DR

A new metric, ensured ranking, is introduced, designed to help users identify the most reliable explanations by balancing trade-offs between uncertainty, probability, and competing alternative explanations.

Abstract

This paper addresses a significant gap in explainable AI: the necessity of interpreting epistemic uncertainty in model explanations. Although current methods mainly focus on explaining predictions, with some including uncertainty, they fail to provide guidance on how to reduce the inherent uncertainty in these predictions. To overcome this challenge, we introduce new types of explanations that specifically target epistemic uncertainty. These include ensured explanations, which highlight feature modifications that can reduce uncertainty, and categorisation of uncertain explanations counter-potential, semi-potential, and super-potential which explore alternative scenarios. Our work emphasises that epistemic uncertainty adds a crucial dimension to explanation quality, demanding evaluation based not only on prediction probability but also on uncertainty reduction. We introduce a new metric, ensured ranking, designed to help users identify the most reliable explanations by balancing trade-offs between uncertainty, probability, and competing alternative explanations. Furthermore, we extend the Calibrated Explanations method, incorporating tools that visualise how changes in feature values impact epistemic uncertainty. This enhancement provides deeper insights into model behaviour, promoting increased interpretability and appropriate trust in scenarios involving uncertain predictions.

Ensured: Explanations for Decreasing the Epistemic Uncertainty in Predictions

TL;DR

A new metric, ensured ranking, is introduced, designed to help users identify the most reliable explanations by balancing trade-offs between uncertainty, probability, and competing alternative explanations.

Abstract

This paper addresses a significant gap in explainable AI: the necessity of interpreting epistemic uncertainty in model explanations. Although current methods mainly focus on explaining predictions, with some including uncertainty, they fail to provide guidance on how to reduce the inherent uncertainty in these predictions. To overcome this challenge, we introduce new types of explanations that specifically target epistemic uncertainty. These include ensured explanations, which highlight feature modifications that can reduce uncertainty, and categorisation of uncertain explanations counter-potential, semi-potential, and super-potential which explore alternative scenarios. Our work emphasises that epistemic uncertainty adds a crucial dimension to explanation quality, demanding evaluation based not only on prediction probability but also on uncertainty reduction. We introduce a new metric, ensured ranking, designed to help users identify the most reliable explanations by balancing trade-offs between uncertainty, probability, and competing alternative explanations. Furthermore, we extend the Calibrated Explanations method, incorporating tools that visualise how changes in feature values impact epistemic uncertainty. This enhancement provides deeper insights into model behaviour, promoting increased interpretability and appropriate trust in scenarios involving uncertain predictions.
Paper Structure (16 sections, 4 equations, 22 figures, 3 tables, 1 algorithm)

This paper contains 16 sections, 4 equations, 22 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Post-Hoc Explanation Methods
  4. Calibration and Uncertainty Quantification
  5. Uncertainty Estimation in Explanations
  6. Calibrated Explanations
  7. Ensured Explanations - Conceptual Framework
  8. Probabilities and Uncertainty
  9. Explanations and Uncertainty
  10. Exploring Alternative Explanations in Probability Space
  11. Ranking Alternative Explanations
  12. Experimental Setup
  13. Results
  14. Experimental Results
  15. Ensured Explanations in Calibrated Explanations
  16. ...and 1 more sections

Figures (22)

  • Figure 1: The structure of post-hoc explanations when considering model prediction.
  • Figure 2: Probabilistic outcome with uncertainty
  • Figure 3: The common structure of post-hoc explanations when adding uncertainty information to model predictions.
  • Figure 4: Exploration of alternatives for probabilities and uncertainty.
  • Figure 5: Probabilistic outcome with uncertainty with counter-, semi- and super-potential explanations
  • ...and 17 more figures