Tag
MUtE introduces a novel framework for optimal concept erasure and counterfactual interventions in text representations, enhancing algorithmic fairness and enabling counterfactual text generation in NLP models.
This paper introduces CDS, a counterfactual directionality score for estimating directional influence between node populations in spatial graphs. The framework trains a Neighbor Influence Model (NIM) and applies structured interventions to measure sensitivity of node states to neighborhood perturbations.