Density-Guided Robust Counterfactual Explanations on Tabular Data under Model Multiplicity
Researchers have developed a new framework called DensityFlow for generating robust counterfactual explanations on tabular data. This method uses a generative approach with Neural ODEs, guided by a density score learned through Noise Contrastive Estimation, to avoid low-density regions where explanations can be unreliable. For black-box models, DensityFlow employs a local proxy distillation mechanism to enable efficient optimization. Experiments show that DensityFlow provides superior validity and reduced query costs compared to existing ensemble-based methods. AI
IMPACT Introduces a novel method for improving the reliability and efficiency of counterfactual explanations in machine learning models.