interpretabilityopinionneutralConcept-based learning offers a promising way to improve model interpretability and generalization but supervised approaches require difficult-to-obtain concept annotationsArtificial Intelligence29 Jul 2026http://arxiv.org/abs/2607.21437v1