REPEAT: Improving Uncertainty Estimation in Representation Learning Explainability