| Creator |
|
|
|
|
|
| Language |
|
| Publisher |
|
| Date |
|
| Source Title |
|
| First Page |
|
| Last Page |
|
| Conference |
|
| Publication Type |
|
| Access Rights |
|
| Related DOI |
|
| Related DOI |
|
| Relation |
|
| Abstract |
Finding linear classifiers that maximize AUC scores is important in ranking research. This is naturally formulated as a 1-norm hard/soft margin optimization problem over pn pairs of p positive and n n...egative instances. However, directly solving the optimization problems is impractical since the problem size (pn) is quadratically larger than the given sample size (p+n). In this paper, we give (approximate) reductions from the problems to hard/soft margin optimization problems of linear size. First, for the hard margin case, we show that the problem is reduced to a hard margin optimization problem over p+n instances in which the bias constant term is to be optimized. Then, for the soft margin case, we show that the problem is approximately reduced to a soft margin optimization problem over p+n instances for which the resulting linear classifier is guaranteed to have a certain margin over pairs.show more
|
Export Mendeley
alt11