CEL: A Unified Library and Benchmark for Counterfactual Explanations in xAI
A new library and benchmark for counterfactual explanations in explainable artificial intelligence (xAI), named CEL (Counterfactual Explanations Library), has been launched by researchers. Counterfactual explanations help identify necessary input modifications to achieve a specific prediction outcome. Initial approaches concentrated on minimal alterations to features, while more recent studies have emphasized aspects like sparsity, actionability, and plausibility. Evaluating these methods fairly has proven difficult due to differing data splits, models, and metrics. CEL resolves this issue by ensuring uniform implementation and assessment. It features 18 datasets with diverse sizes and complexities, along with implementations or reimplementations of 14 popular counterfactual techniques, facilitating objective method comparisons.
Key facts
- CEL stands for Counterfactual Explanations Library.
- It is a unified library and benchmark for counterfactual explanations in xAI.
- Counterfactual explanations provide actionable guidance on input changes to alter model predictions.
- Early methods focused on minimal feature changes.
- Recent work incorporates sparsity, actionability, and plausibility.
- Existing studies often rely on different data splits, predictive models, and evaluation metrics.
- CEL includes 18 datasets of varying size and complexity.
- CEL provides implementations or reimplementations of 14 widely used counterfactual methods.
Entities
—