| D13-1201 | experiments with the gradient | direction finder | of Section 4 are much more positive |
| D13-1201 | and continue with the gradient | direction finder | as soon as the optimum improves |
| D13-1201 | regularized MERT using the gradient | direction finder | and coordinate ascent . At the |
| D13-1201 | the case of the gradient-based | direction finder | , we also use the following strategy |
| D13-1201 | slower due to its lack of a good | direction finder | ) , but our method seems more |
| D13-1201 | Direction finding 4.1 A Gradient-based | direction finder | Perhaps the greatest obstacle |
| D13-1201 | Section 4 , we will present a novel | direction finder | for maximum-BLEU optimization |
| D13-1201 | regularized MERT with the gradient | direction finder | took 37 minutes on average , |
| D13-1201 | are much more positive . This | direction finder | not only recovers w * ( cosine |
| D11-1004 | guarantee : without a perfect | direction finder | , even a globally-exact line |
| H90-1044 | it is formatted and sent to the | direction finder | . When the direction finder 's |
| H90-1044 | user to refer to things that the | direction finder | has mentioned , and in general |
| D13-1201 | thanks to regularization and a | direction finder | that directs the search towards |
| H90-1044 | the direction finder . When the | direction finder | 's English response is returned |
| H90-1044 | direction finder ( QTIP ) , and the | direction finder | itself . VFE takes SUMMIT 'S |
| H90-1044 | that the information from the | direction finder | 's response can be processed |
| D13-1201 | regularization and the gradient-based | direction finder | . All variants of MERT are initialized |
| H90-1044 | formats PUNDIT 's output for the | direction finder | ( QTIP ) , and the direction |
