Lilt：以“Weekly Summary 1. New round of experiments have done to prove the effectiveness of scoring domain feature. Three group experiments have done and two of these show...”为内容创建页面

2014-11-02T09:55:20Z

以“Weekly Summary 1. New round of experiments have done to prove the effectiveness of scoring domain feature. Three group experiments have done and two of these show...”为内容创建页面

新页面

Weekly Summary

1. New round of experiments have done to prove the effectiveness of scoring domain feature.

Three group experiments have done and two of these show positive results: EER of baseline > Tnorm > Fusion > Total.

The different between Fusion and Total is the SVM training data. Fusion is the score of fuzzy range. Total is the total score.

The scoring feature of each is system score(1) / Tnorm score(1) / score + delta score + Tnorm score(22) / score + delta score + Tnorm score(22).

Using Total data can increase the amount of training data and improve the system performance.

2. Try to adjust parameters C(penalty) to solve the 'rbf' overfitting problem, setting C = 0.3/0.6/1(baseline)/5/10, but results are still so bad.

3. To help Prof.Zheng revise APSIPA 2014 paper and submit it.

Next Week

1. Submit other APSIPA paper files.

2. Go on to solve overfitting problem.

3. Apply this score domain method to a new dataset.

Lantian Li 14-11-03 - 版本历史

Lilt：以“Weekly Summary 1. New round of experiments have done to prove the effectiveness of scoring domain feature. Three group experiments have done and two of these show...”为内容创建页面