Towards Assumption-free Bias Mitigation