| Abstract | Our formulation is able to handle nonlocal output features in an efficient manner; not only is it compatible with prior knowledge encoded as hard constraints, it can also learn soft constraints from data. |
| Conclusions | These features can act as soft constraints whose penalty values are automatically learned from data; in addition, our model is also compatible with expert knowledge in the form of hard constraints. |
| Introduction | 0 Soft constraints may be automatically learned from data. |