The author proposes ERASE, a method for learning error-resilient representations on graphs to enhance robustness against label noise in deep learning models.
