All Notes Project Overview
Updated April 1, 2025

Lime

LIME is a technique within Explainable AI (XAI) designed to explain the predictions of any machine learning classifier or regressor in an interpretable manner. It works by approximating the complex model locally around a specific prediction using a simpler, interpretable model (like linear regression or a decision tree).

How it Works

  1. Perturbation: LIME generates variations of the input instance for which an explanation is sought (e.g., slightly changing words in text, turning pixels on/off in images).
  2. Prediction: It gets predictions from the original complex (“black box”) model for these perturbed instances.
  3. Local Approximation: LIME trains a simple, interpretable model (e.g., weighted linear regression) on these perturbed instances and their predictions, weighting instances closer to the original input more heavily.
  4. Explanation: The interpretable model’s parameters (e.g., feature weights in the linear model) are used as the explanation for the original prediction. For text, this often highlights words contributing positively or negatively to a classification; for images, it highlights influential superpixels.

Key Characteristics

  • Model-Agnostic: Can be applied to any model without needing internal access.
  • Local Fidelity: Aims to explain individual predictions accurately in their local vicinity, rather than global model behavior.
  • Interpretability: Provides explanations using simpler models that are easier for humans to understand.

Use Cases

  • Explaining specific classification decisions (e.g., why an email was marked as spam).
  • Identifying key features driving a prediction for debugging or user feedback.
  • Building trust by showing the basis for an AI recommendation.

Connections

References