LIPEx-Locally Interpretable Probabilistic Explanations-To Look Beyond The True Class

LIPEx provides multi-class explanations, not just single-class: Unlike some existing methods, LIPEx shows how features influence prediction probabilities across all possible classes, not just the predicted one. This helps users understand why a model might be uncertain between multiple options or what would need to change for it to predict a different class.

More efficient than current methods: LIPEx requires fewer data perturbations (variations) than LIME, a widely-used explanation tool, to produce reliable explanations—and does so approximately 53% faster for text classification. This efficiency matters for practical deployment where computational resources are limited.

Better at identifying truly important features: When researchers removed features that LIPEx identified as important, the model's predictions changed more dramatically than when removing features identified by other explanation methods. This suggests LIPEx more accurately pinpoints which features actually drive model decisions.

Works across different data types: The method was tested on both text and image data with consistent results, indicating it could be broadly applicable across different AI systems rather than being limited to one domain.

Addresses a gap in explainability: By explaining probability distributions rather than just final predictions, LIPEx helps users understand model confidence and uncertainty—critical information for responsible AI deployment in high-stakes decisions where knowing "how sure" the model is matters as much as what it predicts.