If you were given a dataset with biased labels, how would you address this issue in your machine learning model?