L2X
A method for selecting interpretable feature subsets from complex models using mutual information optimization.
What is L2X?
L2X is a model interpretation method that selects the most informative feature subsets to explain predictions from complex machine learning models. It formulates feature selection as an optimization problem that maximizes mutual information between selected features and model outputs. The approach provides instance-specific explanations while maintaining the original model's predictive accuracy.
Machine learning researchers and practitioners working on explainable AI who need to interpret black-box model predictions, particularly those using deep learning models where traditional interpretation methods are insufficient.
L2X offers an information-theoretic foundation for model interpretation that provides mathematically grounded feature importance scores. Unlike gradient-based methods, it selects discrete feature subsets and works with any black-box model without requiring access to gradients or internal parameters.
Overview
L2X (Learning to Explain) is a model interpretation framework that identifies the most informative features for a model's predictions through an information-theoretic approach. It provides instance-wise feature selection to explain black-box models while maintaining predictive performance.
Key Features
- Mutual Information Optimization — Selects feature subsets that maximize mutual information with the target variable
- Instance-Wise Explanations — Provides explanations tailored to individual predictions rather than global model behavior
- Model-Agnostic Framework — Works with various black-box models without requiring access to internal parameters
- Interpretable Feature Selection — Identifies compact subsets of features that are most relevant for each prediction
Philosophy
L2X approaches model interpretation as an information-theoretic optimization problem, selecting minimal feature subsets that preserve maximal information about model predictions.
Related Projects
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
