Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Which LIME should I trust? Concepts, Challenges, and Solutions

Patrick Knab, Sascha Marton, Udo Schlegel, Christian Bartelt

TL;DR

This work surveys LIME and its proliferating variants, addressing fidelity, stability, and domain-specific challenges. It proposes a two-dimensional taxonomy of LIME extensions that separates technical modifications from the issues they target, and maps these across modalities and domains. The authors provide methodological guidance, propose evaluation and reproducibility considerations, and offer a continuously updated web resource to track developments. Together, these contributions help researchers and practitioners select appropriate LIME approaches and identify promising directions for future improvements in explainable AI.

Abstract

As neural networks become dominant in essential systems, Explainable Artificial Intelligence (XAI) plays a crucial role in fostering trust and detecting potential misbehavior of opaque models. LIME (Local Interpretable Model-agnostic Explanations) is among the most prominent model-agnostic approaches, generating explanations by approximating the behavior of black-box models around specific instances. Despite its popularity, LIME faces challenges related to fidelity, stability, and applicability to domain-specific problems. Numerous adaptations and enhancements have been proposed to address these issues, but the growing number of developments can be overwhelming, complicating efforts to navigate LIME-related research. To the best of our knowledge, this is the first survey to comprehensively explore and collect LIME's foundational concepts and known limitations. We categorize and compare its various enhancements, offering a structured taxonomy based on intermediate steps and key issues. Our analysis provides a holistic overview of advancements in LIME, guiding future research and helping practitioners identify suitable approaches. Additionally, we provide a continuously updated interactive website (https://patrick-knab.github.io/which-lime-to-trust/), offering a concise and accessible overview of the survey.

Which LIME should I trust? Concepts, Challenges, and Solutions

TL;DR

This work surveys LIME and its proliferating variants, addressing fidelity, stability, and domain-specific challenges. It proposes a two-dimensional taxonomy of LIME extensions that separates technical modifications from the issues they target, and maps these across modalities and domains. The authors provide methodological guidance, propose evaluation and reproducibility considerations, and offer a continuously updated web resource to track developments. Together, these contributions help researchers and practitioners select appropriate LIME approaches and identify promising directions for future improvements in explainable AI.

Abstract

As neural networks become dominant in essential systems, Explainable Artificial Intelligence (XAI) plays a crucial role in fostering trust and detecting potential misbehavior of opaque models. LIME (Local Interpretable Model-agnostic Explanations) is among the most prominent model-agnostic approaches, generating explanations by approximating the behavior of black-box models around specific instances. Despite its popularity, LIME faces challenges related to fidelity, stability, and applicability to domain-specific problems. Numerous adaptations and enhancements have been proposed to address these issues, but the growing number of developments can be overwhelming, complicating efforts to navigate LIME-related research. To the best of our knowledge, this is the first survey to comprehensively explore and collect LIME's foundational concepts and known limitations. We categorize and compare its various enhancements, offering a structured taxonomy based on intermediate steps and key issues. Our analysis provides a holistic overview of advancements in LIME, guiding future research and helping practitioners identify suitable approaches. Additionally, we provide a continuously updated interactive website (https://patrick-knab.github.io/which-lime-to-trust/), offering a concise and accessible overview of the survey.

Paper Structure

This paper contains 13 sections, 3 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Contribution and Organization
  3. XAI Categorization
  4. Methodology
  5. Fundamentals of LIME
  6. LIME Categorization
  7. Issue Categorization
  8. Implementation Categorization
  9. Domain Categorization
  10. Discussion, Opportunities & Conclusion
  11. Discussion
  12. Research Opportunities
  13. Conclusion

Figures (3)

  • Figure 1: https://patrick-knab.github.io/which-lime-to-trust/. This website is designed to monitor and collect new LIME-related techniques continuously. For an ongoing collection of LIME-related methods, please refer to the webpage: https://patrick-knab.github.io/which-lime-to-trust/.
  • Figure 2: Steps of LIME: The framework operates in four steps: (1) Feature generation: Extract features (e.g., image segmentation). (2) Sample generation: Create perturbed samples around the instance. (3) Feature attribution: Train an interpretable model (e.g., linear) to approximate the complex model locally. (4) Explanation representation: Use the model’s weights to represent feature importance.
  • Figure 3: LIME Exemplary Explanations.\ref{['fig:text']} shows a sentiment classification from a movie review (IMDB dataset maas-EtAl:2011:ACL-HLT2011), highlighting words linked to positivity and negativity. \ref{['fig:tab']} depicts a young female passenger from the Titanic dataset, with survival probability mainly influenced by her sex and age. \ref{['fig:image']} explains an image classified as a gorilla, where green and red superpixels represent positive and negative classification contributions.