Injecting fairness into machine-learning models

MIT researchers have found that, if a certain type of machine learning model is trained using an unbalanced dataset, the bias that it learns is impossible to fix after the fact. They developed a technique that induces fairness directly into the model, no matter how unbalanced the training dataset was, which can boost the model’s performance on downstream tasks.

Related Keywords

Toronto ,Ontario ,Canada ,Natalie Dullerud ,Nicolas Papernot ,Marzyeh Ghassemi ,Karsten Roth ,Kimia Hamidieh ,International Conference On Learning Representations ,Artificial Intelligence Laboratory ,Healthy Ml Group Of The Computer Science ,University Of Tubingen ,University Of Toronto Department Electrical Engineering ,Healthy Ml Group ,Computer Science ,Electrical Engineering ,International Conference ,Partial Attribute Decorrelation ,Deep Metric Learning ,Airness In Artificial Intelligence ,