Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Causality-Aware Local Interpretable Model-Agnostic Explanations

Martina Cinquini, Riccardo Guidotti

TL;DR

This paper proposes a novel extension to a widely used local and model-agnostic explainer, which encodes explicit causal relationships within the data surrounding the instance being explained and overcomes the original method in terms of faithfully replicating the black-box model's mechanism and the consistency and reliability of the generated explanations.

Abstract

A main drawback of eXplainable Artificial Intelligence (XAI) approaches is the feature independence assumption, hindering the study of potential variable dependencies. This leads to approximating black box behaviors by analyzing the effects on randomly generated feature values that may rarely occur in the original samples. This paper addresses this issue by integrating causal knowledge in an XAI method to enhance transparency and enable users to assess the quality of the generated explanations. Specifically, we propose a novel extension to a widely used local and model-agnostic explainer, which encodes explicit causal relationships within the data surrounding the instance being explained. Extensive experiments show that our approach overcomes the original method in terms of faithfully replicating the black-box model's mechanism and the consistency and reliability of the generated explanations.

Causality-Aware Local Interpretable Model-Agnostic Explanations

TL;DR

This paper proposes a novel extension to a widely used local and model-agnostic explainer, which encodes explicit causal relationships within the data surrounding the instance being explained and overcomes the original method in terms of faithfully replicating the black-box model's mechanism and the consistency and reliability of the generated explanations.

Abstract

A main drawback of eXplainable Artificial Intelligence (XAI) approaches is the feature independence assumption, hindering the study of potential variable dependencies. This leads to approximating black box behaviors by analyzing the effects on randomly generated feature values that may rarely occur in the original samples. This paper addresses this issue by integrating causal knowledge in an XAI method to enhance transparency and enable users to assess the quality of the generated explanations. Specifically, we propose a novel extension to a widely used local and model-agnostic explainer, which encodes explicit causal relationships within the data surrounding the instance being explained. Extensive experiments show that our approach overcomes the original method in terms of faithfully replicating the black-box model's mechanism and the consistency and reliability of the generated explanations.
Paper Structure (10 sections, 3 equations, 5 figures, 2 tables)

This paper contains 10 sections, 3 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Background
  4. Causality-Aware LIME
  5. Experiments
  6. Datasets and Classifiers
  7. Comparison with Related Works
  8. Evaluation Measures
  9. Results
  10. Conclusion

Figures (5)

  • Figure 1: The high-level workflow of our framework, with the blue dashed line showing the difference in generating the synthetic neighborhood compared to lime.
  • Figure 2: (Left). Feature importance computed by calime and lime. (Right). Causal graph of banknote inferred by a causal discovery method hoyer2008nonlinear.
  • Figure 3: Fidelity as $\boldsymbol{R^2}$ varying the number of features for statlog, wdbc and wine-red. Markers represent the mean values, while vertical bars are the standard deviations.
  • Figure 4: Plausibility errors as $AMD$, $AOD$, $ASM$, and $ADM$ in box plots aggregating the results across the scores obtained with different numbers of features. Best viewed in color.
  • Figure 5: Instability as $\boldsymbol{LLE}$ varying the number of features. Markers represent the mean values, while the contingency area highlights the minimum and maximum values.