δ-CLUE: Diverse Sets of Explanations for Uncertainty Estimates
This work addresses the need for interpretable uncertainty explanations in machine learning, offering a more comprehensive approach for users who rely on model confidence, though it is incremental as it builds on existing CLUE methods.
The paper tackles the problem of generating diverse explanations for uncertainty estimates in probabilistic models by introducing δ-CLUE, which produces multiple, diverse counterfactual explanations within a δ ball in latent space to increase model confidence, resulting in a method that outputs sets of plausible explanations rather than a single one.
To interpret uncertainty estimates from differentiable probabilistic models, recent work has proposed generating Counterfactual Latent Uncertainty Explanations (CLUEs). However, for a single input, such approaches could output a variety of explanations due to the lack of constraints placed on the explanation. Here we augment the original CLUE approach, to provide what we call $δ$-CLUE. CLUE indicates $\it{one}$ way to change an input, while remaining on the data manifold, such that the model becomes more confident about its prediction. We instead return a $\it{set}$ of plausible CLUEs: multiple, diverse inputs that are within a $δ$ ball of the original input in latent space, all yielding confident predictions.