Relax and penalize: a new bilevel approach to mixed-binary hyperparameter optimization

Open in new window